Post on 16-Oct-2020
Liquidity and Capital Structure
Ronald W. Anderson ∗and Andrew Carverhill †
February 9, 2010
Abstract
We solve for a firm’s optimal cash holding policy within a continuous time, con-tingent claims framework using dividends, short-term borrowing and equity issues ascontrols. Under the optimal policy the firm targets a level of cash holding that is anon-monotonic function of business conditions and an increasing function of the levelof long-term debt. The model matches closely a wide range of empirical benchmarksand predicts cash and leverage dynamics in line with the empirical literature. Firmvalue is largely insensitive to the level of long-term debt. The optimal policy exhibitsa state-dependent hierarchy that agrees with recent explorations of pecking order the-ory. In an extension we find that targeted cash holdings are not much increased by thepresence of growth opportunities until shortly before the growth occurs. We show thatobserved bond covenants that establish an earnings restriction on dividend paymentsmay be value increasing.
∗London School of Economics, r.w.anderson@lse.ac.uk†Hong Kong University, carverhill@business.hku.hk
Liquidity and Capital Structure
“Kerkorian’s numbers just don’t add up,” said Nicholas Lobaccaro, an auto an-
alyst with S G Warburg. “Ford says it needs double-digit billions of cash to
survive the next downturn in the market. General Motors says it wants to put
aside $13-15 billion. How can anyone believe Kerkorian when he says $2 billion
is enough?”[for Chrysler] 1
1 Introduction
The quotation above illustrates the range of opinions that can be found among practitioners
about the levels of liquid assets that are appropriate for firms. This observation is not
an isolated case - it is often remarked that many large corporations carry surprisingly large
amounts of cash on their balance sheets. However, finance theory has given very little precise
guidance as to how much cash is enough.
In this paper we attempt to fill this gap by directly asking what is the optimal policy
toward holding liquid assets in the firm. Specifically we determine the dynamically optimal
cash holding and leverage policy in a firm with given assets in place and long-term debt
outstanding. The controls are the firm’s dividend, short-term borrowing, and share issuance
choices. This analysis is important because in standard finance teaching cash holding is
treated as a short-term, operational problem separate from the firm’s capital structure choice.
In fact, for a firm choosing cash holding with foresight, this choice is inherently related to
the long-term capital structure. Under the optimal policy the level of cash is adjusted
dynamically as the firm is exposed to changing business conditions. We characterize the
qualitative features of this policy. In addition, because our model employs a relatively rich
continuous time formulation of the firm’s cash flow process, we can calibrate the model to
several empirical benchmarks and thereby gain some insight into the quantitative magnitudes
of optimal cash holding in a variety of circumstances.
The core qualitative finding is that the optimal cash holding of the firm is a hump-shaped
function of the firm’s expected cash flow with the peak average cash holding occurring in
the neighborhood of the firm’s long-term rate of cash flow. As business conditions improve
1Sunday Times of London, April 23, 1995. This quote refers to the attempt by Kirk Kekorian to take overChrysler Motors arguing that in doing so he could increase shareholder value by returning most of Chrysler’s$7.5 billon cash reserve to shareholders.
1
beyond the long-term average, the firm targets a lower level of cash and pays dividends that
maintain cash at the target level. Depending upon the firm’s level of long-term debt, it may
be optimal for the firm to borrow short-term in order to increase leverage. When business
conditions fall below the long-term rate, the firm will target a higher level of cash holdings
and will suspend dividends in an attempt to achieve this. However, with the decline in
revenues, the firm will see average cash holdings fall well below the target level. When there
is a sustained or deep decline in revenues, cash will drop to a minimum level at which point
the firm will issue equity. However, at a still lower threshold of expected cash flow, the firm’s
equity value drops to zero at which point the firm is bankrupt and the remaining assets are
transferred to creditors.
A rich variety of consequences follow from this optimal policy. Perhaps the most im-
portant is that since the firm is able to control its net leverage through its dynamic cash
policy, firm value is relatively insensitive to its long-term debt level structure. That is, in
contrast with static trade-off models where the central issue is the choice of capital structure
in the face of tax benefits of debt and bankruptcy costs, under the dynamically optimal
cash and dividend policy, the choice of long-term debt levels is of secondary importance.
Another important implication is that there is a hierarchy or “pecking order” among sources
of funds but this hierarchy is a function of business conditions. If business conditions are
very good the hierarchy is internal funds first, short-term debt second, and equity third.
However, in poorer business conditions the firm will avoid short-term debt and will turn to
equity issuance if internal funds are not available. A third, implication is that in contrast
with the basic corporate finance teaching about the “asset substitution” problem, for a firm
that chooses cash dynamically to maximize firm value, the interests of debt holders and
shareholders are aligned with respect to the choice of asset volatility.
We explore the empirical implications of the model in more detail by simulating the
model in the baseline case calibrated to match simultaneously benchmarks for cash holdings,
leverage, equity volatility, yield spreads, default probabilities and recovery rates. Using
sample paths generated from this baseline model, we are able obtain results very similar
to the cross-sectional and time series results reported in recent empirical studies of cash
holding. When the model is extended to allow for an increase in the scale of the firm’s risky
technology, we find that the growth opportunity does not greatly impact cash holdings until
shortly before the growth takes place. We further use the model to calculate the agency
costs generated by the conflict of interest between shareholders and creditors regarding the
firm’s liquidity policy and show that bond covenants that establish an earnings restriction
on dividend payments may be value increasing.
The papers most closely related to ours are Mello and Parsons (2000), Hennessy and
2
Whited (2005) and Rochet and Villeneuve (2004), all of which have provided theoretical
treatments of the firm’s financial policy which determine liquidity holdings. The models
of these papers share some but not all of the features of the present paper. Hennessy and
Whited consider a discrete time model with cash and short-term debt but no long-term debt.
They are mainly concerned with the dynamic capital structure implications of the analysis.
The papers by Mello and Parsons and by Rochet and Villeneuve are concerned with the
benefits of hedging. A detailed discussion of the relation of our model and findings to the
literature will be given once we have presented our model.
The remainder of the paper is organized as follows. In Section 2 we introduce the model
and the dynamic programming technique we use to solve it. We explore in some detail a
benchmark solution of the model and discuss its relation to previous empirical findings. In
Section 3 we draw out the implications of our model for the firm’s choice of capital structure.
We also study the consequences for firm behavior of changes of the firm’s technology and
of contracting frictions of the economic environment where the firm operates. Section 4
is devoted to two extensions of the basic model. First we study the implication of fixed
investment for cash holding. Second, we analyze the agency conflict associated with the
liquidity reserve and consider debt covenants intended to mitigate this conflict. In Section 5
we summarize our results and conclusions. Finally, in an Appendix we present some technical
details of our numerical procedure.
2 The Model
2.1 Overview
Before presenting our model formally, it is useful to set out the main ideas in informal
terms. We consider a firm with a fixed asset in place which has been financed by equity,
variable short-term debt and fixed long-term debt. The asset generates a random cash flow
according to a stochastic process whose drift is itself random and follows a mean-reverting
process. Any cash flow in excess of contractual debt service and fixed operating costs is
subject to proportional corporate income tax, and the after-tax residual may either be paid
out as dividends, used to reduce short term debt, or retained as liquid assets within the firm.
Debt is assumed to be a hard claim, and any failure to meet contractual debt service results
in bankruptcy. We assume that strict priority is observed in bankruptcy, with the firm’s
assets in excess of bankruptcy costs being awarded to the firm’s creditors. Shareholders lose
all. When cash flows fall short of debt service, the firm may draw-down its liquid assets or
issue short term debt. It may also issue new equity; however, this external finance is costly
3
so that the firm receives less than the full value of the shares it issues. The asset in place is
indivisible so that partial sales of the risky asset are not allowed.
In this setting, firm faces two decisions. How much of the firm’s earnings should be paid
out as dividends? And how many new shares should be issued? Jointly, the two decisions
will determine the firm’s policy toward holding liquid assets and short term debt issuance.
In our framework there is no reason to hold cash and borrow short term simultaneously.
We assume that these decisions are under the control of shareholders, who maximize the
value of equity, calculated as the present discounted value of the future stream of dividends.
The firm’s decision will depend upon two state variables: the current rate of revenue cash
flow and the current level of liquid assets (which is interpreted as borrowing when negative).
Since all the other features of the environment are constant, this is a stationary problem.
The solution of the model involves solving for the optimal policy as a function of the two
state variables.
Shareholders face different costs for alternative capital market operations. Firm insiders
will to some degree extract rents from liquid assets inside the firm, so that inside cash will
grow at something less than the money market rate. On the other hand, the firm borrows at
a rate greater than the money market rate reflecting informational rents conceded as part of
its banking relationship. Finally, issuing equity will incur floatation costs. In this context,
the optimal dividend and share issuance strategy will be of the ‘bang-bang’ type, under
which the state space is divided into 3 regions: in the ‘save’ region zero dividend is paid
and earnings are accumulated in the reserve of liquid assets or used to pay down short term
debt; in the ‘dividend’ region, the liquid reserve is immediately paid out, until it is brought
back to the ‘save’ region, or to abandonment or bankruptcy; and in the ‘issue’ region, the
firm immediately issues equity until liquid reserve is brought back into the ‘save’ region. The
solution to the problem is studied by characterizing the boundaries between these regions as
free boundaries, in a dynamic program.
2.2 Model Assumptions and Detailed Specification
The firm has fixed assets in place2, which incur operating costs at a constant rate f , and
which generate operating revenues at a rate dSt according to the Ito equation
dSt = ρtdt + σdW σt . (1)
Here expected revenue ρt at time t itself obeys the Ito equation
dρt = κ(ρ− ρt)dt +√
ρtηdW ρt . (2)
2In section 4 we relax this assumption by allowing for growth opportunities.
4
In these equations, dW σt and dW ρ
t are infinitesimal increments of independent, standard
Brownian motions, and κ, ρ, σ and η are positive constants. Equation (2) reflects our as-
sumption that the expected rate of revenue is positive and mean reverting, representing
variation of business conditions over the business cycle3. Mean reversion of ρt is discussed
further in Section 3.3 below. Equation (1) adds an unpredictable random element to the
operating income. Note that after deducting the operating costs the profitability of the fixed
assets is given by dSt − fdt.
The firm is financed by equity, variable short-term debt, and fixed long-term debt.4 We
assume that the firm’s short-term debt is instantaneously maturing, and long-term debt takes
the form of a perpetual bond promising a continuous payment at rate q. In practice, a typical
financial structure will involve short-term secured bank loans and long-term debentures which
may be protected against dilution through restrictions on the amount of debt issuance (see,
Petersen & Rajan, 1994, and Bolton & Freixas, 2000). We capture this by assuming that
given a fixed q the firm can instantaneously borrow up to its debt capacity, determined
in our model by the value of the firm in bankruptcy. This will have the effect of making
short-term debt default risk-free. We assume that the firm pays an interest rate on short-
term borrowing, rbank, which is in excess of the risk-free money market rate, r. The amount
rbank−r reflects the informational rents ceded by the firm as part of its banking relationship.
The firm also may issue new equity to cover interest payments and operating losses. Such
equity issues will be costly, in that the firm will be able to sell new shares at a fraction θ < 1
of their fair value. The parameter θ will reflect fees and pricing concessions associated with
primary equity market operations and may vary systematically with the efficiency of the
capital markets where the firm operates. In a highly efficient market θ will be close to unity;
whereas in a very underdeveloped capital market θ may be close to zero.
In addition to its fixed asset, the firm may hold a variable amount of liquid reserves. At
any time t, the value of these will be denoted by Ct. Liquid reserves held within the firm
will earn an ‘internal’ return at rate rin, which will be less than the riskless rate r earned
on outside funds. This wedge r − rinbetween rin and r reflects the moral hazard faced by
the shareholders, as discussed by Myers and Rajan (1998). As already mentioned, in our
framework there will be no incentive to hold cash and issue short-term debt simultaneously.
3This equation also has the property that varying the volatility η does not alter the expectation of ρs forgiven ρt with t < s. See Duffie (2001). This will be useful when we discuss the asset substitution effect inour model.
4As in standard trade-off models the tax deductibility of interest payments provides an incentive to issuedebt. Recently, DeMarzo and Sannikov (2006) have studied the problem of security design in a continuoustime model without taxes and find that the optimal capital structure involves a fixed amount of long-timedebt and varying short-term debt as in our model.
5
Thus Ct < 0 will correspond to a situation where the firm is borrowing short-term.
Under these assumptions, and for the time-being ignoring the possibility of equity issues,
the liquid reserve is the accumulation of total earnings net of dividends, fixed costs, and
interest payments on debt, and we can write
dCt = (1− τ)(dSt − ((f + q)− rCtCt)dt)− dDt, (3)
where τ is the corporate tax rate, and Dt is the accumulated payments of dividends, rC
is equal to rin when C > 0, and equal to rbank when C ≤ 0. This equation recognizes
that tax is paid on the operating income and interest on the cash reserve, net of interest
payments and fixed costs, i.e. dSt − ((f + q)− rCtCt)dt. Applying this to negative earnings
is an analytically tractable way to model the loss carry-back and carry-forward provisions of
many tax regimes.5
The firm becomes bankrupt if it does not meet its debt obligations and fixed costs, either
from its operating revenue, its cash reserve, or by issuing new shares. On bankruptcy, the
creditors are awarded the firm’s fixed assets. We assume their value equals the value of the
unlevered firm less bankruptcy costs, which we take to be a fraction α of this value. We
denote the equity value with long-term debt coupon, q, by Jq(ρ, C). Then the debt holders
get value (1−α)J0(ρ, C) upon bankruptcy. We assume that the firm can borrow short-term
up to the value of this collateral less the face value of long-term debt, q/r. Thus for given
ρ, the short-term borrowing limit will be a negative cash holding denoted C(ρ), given by
C(ρ) = −min {0, (1− α)J0(ρ, C)− q/r}Finally, we assume the firm chooses the dividend and capital market policy so as to
maximize equity value, which is taken to be the present value of expected dividends dis-
counted at the risk-free rate r. The debt is also valued by discounting at the risk free rate
the coupon payments until bankruptcy, and then the bankruptcy value. This is consistent
with Equations (1), (2) referring to the risk neutral probability measure.6
5This is similar to the treatment of corporate taxes in Hennessey and Whited (2005). They model taxwrite-back and carry-forward provisions by allowing the tax rate to be progressive in terms of earnings.However, their estimated parameters imply that the marginal tax rate is essentially constant over the eco-nomically relevant range.
6Note that in our formulation bankruptcy occurs when our state variable evolving along a continuoussample path attains an absorbing barrier. Thus unlike reduced-form credit risk models there is no “jump-to-default risk”, e.g., as in Duffie and Singleton (1999). They discount cash flows at the rate r + ξt, whereξt is the instantaneous hazard rate of default. In this terminology, our model assumes ξt = 0.
6
2.3 PDEs for the Solution
Under the above assumptions, and ignoring for the moment the possibility of new equity
issues, the value of the firm’s equity is determined at any time t by the current values of
profitability ρ and cash reserve C, which represents short term borrowing, if it is negative.
Denoting this value by Jqt (ρ, C), then we can write the HJB equation
Jqt (ρ, C) = max
dDt
{dDt + e−rdtE
(ρ,C)t
[Jq
t+dt(ρt+dt, Ct+dt)]}
, (4)
in which dt is an infinitesimally short time step and dDt is the optimal dividend payment
over this time step, which must be non-negative. Also, E(ρ,C)t means the expectation at time
t, given that (ρt, Ct) = (ρ, C). If the liquid reserve becomes low, then the firm can increase
it by issuing more equity, if this is feasible in terms of the share price.
Expanding Jqt+dt(ρt+dt, Ct+dt) in Equation (4), using the Ito Formula, E[dW σ
t ] = 0, and
following standard manipulations we obtain the Ito Equation
Jqt (ρ, C)(1− e−rdt) = max
dDt≥0
{dDt +
(∂
∂tJq
t + κ(ρ− ρ)∂
∂ρJq
t +1
2ρη2 ∂2
∂ρ2Jq
t
)dt
+ [(1− τ)(ρ− (f + q) + rCC)dt− dDt]∂
∂CJq
t +1
2σ2(1− τ)2 ∂2
∂C2Jq
t dt
}(5)
We emphasize that this equation holds only for the optimal choice of dDt, which depends
on (ρ, C).
The optimal choice of dDt here is singular: if ∂∂C
Jqt < 1, then it is optimal to pay
dividends as quickly as possible, reducing the cash holding until either ∂∂C
Jqt ≥ 1, or until
the firm becomes bankrupt. If ∂∂C
Jqt ≥ 1, then the firm will not pay dividends.
The optimal decision can thus be characterized in terms a “save” region S and a “div-
idend” region D in the state space {(ρ, C) : C ≥ C(ρ)}. In S we have ∂∂C
Jqt > 1, and also
Equation (5) holds, with dDt/dt = 0, i.e.
∂
∂tJq
t − rJqt + κ(ρ− ρ)
∂
∂ρJq
t +1
2ρη2 ∂2
∂ρ2Jq
t
+ [(1− τ)(ρ− (f + q)) + rCC]∂
∂CJq
t +1
2σ2(1− τ)2 ∂2
∂C2Jq
t = 0. (6)
In D we have∂
∂CJq
t = 1, (7)
and Equation (6) does not apply, since the value of an extra dollar in the liquidity reserve
is just its value if immediately paid as a dividend. If the liquid reserve Ct becomes so high
7
that (ρt, Ct)εD, then a dividend should immediately be paid, to take (ρt, Ct) back into the
region S , or to bankruptcy.
If the liquid reserve becomes low, then it may be optimal for the firm to issue new equity.
We have not included this possibility in the above formulation. In fact it is optimal to issue
more equity if 7 ∂∂C
Jqt > 1
θ. This possibility leads to there being a third, ‘issue’ region, which
we will denote by I, lying below S, and in which
∂
∂CJq
t =1
θ. (8)
If the liquid reserve Ct becomes low, so that (ρt, Ct)εI, then new equity should immediately
be issued, to take (ρt, Ct) back into the region S . Note that until bankruptcy occurs, the
process (ρt, Ct) will always remain in the save region S, since it is immediately pushed away,
whenever it enters the region D or I.
These regions must be chosen to maximize Jqt (ρ, C), which implies that there must be
value matching and smooth pasting of the solution across the boundaries between S, D, and
I. Also, these boundaries are ‘free’, in that they are determined as part of the solution to
Equations (6), (7), (8) with value matching and smooth pasting. The boundary condition at
the lower boundary {C(ρ)}ρ is just Jq(ρ, C(ρ)) ≥ 0. Our solution technique will be described
in detail in the Appendix below. The approach is to develop the solution to Equation (6)
numerically from a distant horizon t = T to t = 0, and to test at every point whether the
value is increased by applying Equation (7) or (8), instead of Equation (6). The horizon T
is taken to be sufficiently far away, that the solution is independent of T for t near zero.
The debt value P qt (ρ, C) can be calculated by solving a PDE, in a similar way to the
equity value Jqt (ρ, C) above. In fact, the calculation is simpler because the boundaries of
the region in which the debt is defined, i.e. S, have already been determined in the equity
valuation. The PDE for the debt, in the region S, is
q +∂
∂tP q
t − rP qt + κ(ρ− ρ)
∂
∂ρP q
t +1
2ρη2 ∂
∂ρ2P q
t +
[(1− τ)(ρ− (f + q) + rCC)]∂
∂CP q
t +1
2σ2 ∂
∂C2P q
t = 0. (9)
7Proof: Suppose the current cash holding C is too low, and the firm raises δC in an equity issue, toincrease the cash holding. Suppose the firm initially has N shares, and issues n more shares. Denote bys the share price after issue. Then the obtained from selling each new share is θs, and also n = δC/θs
and s = J(C + δC)/(N + n). These imply that s = [J(C + δC) − δC/θ]/N . Now, the firm will issueshares if it increases the share price. Before issue, the share price is J(C)/N , and so the firm will issue ifJ(C)/N < [J(C + δC)− δC/θ]/N , which implies that [J(C + δC)− J(C)]/δC > 1/θ. QED
8
Again, we evolve the solution backwards from a distant horizon t = T . The boundary
conditions for the debt valuation are as follows: First ∂∂C
P qt = 0, where S meets I or where
S meets D and D is above S. This corresponds to the reflection of the process Ct at these
boundaries8; Second P qt = (1 − α)J0
0 (ρ, 0) where S meets D and D is below S, or when
C = C(ρ) and ρ is not high enough to induce the equity holders to maintain payments to
debt and fixed costs. This corresponds to bankruptcy, under which the debt holders receive
the unlevered value of the firm, net of bankruptcy costs. (Note that the restrictions on short
term debt mean that the firm will not go bankrupt if it has any short term debt, and so we
must have C = C(ρ) = 0 at bankruptcy.)
2.4 Benchmark Solutions to the PDEs
We take our benchmark parameter set to be rin = 4%, r = 6%, rbank = 8%, f = 0.14, ρ =
0.15, η = 0.09, κ = 0.9, σ = 0.0, τ = 30%, θ = 0.8, α = 0.3, and q = 0.004. Some of
these parameters have a direct economic interpretation. Notice that by setting r = 6% and
rin = 4% we are assuming that one third of the market return on cash is dissipated by
keeping the cash inside the firm and under the control of management. We view this as a
reasonably severe problem of managerial moral hazard and a rather strong disincentive to
holding cash. In this sense, the levels of cash holding our model predicts might be viewed
as conservative. Similarly, setting rbank = 8%, we are assuming a significant relationship
premium on short-term borrowing.9 By setting θ = 0.8 we assume 20% of the market value
of newly issued equity is lost through transactions costs of one form or another. Given the
direct costs plus underpricing of equity issues, we view these costs as substantial but not
unreasonable in many settings.10 Our assumption of bankruptcy costs of 30% at the high end
of estimates that can be found in the literature.11 We explore the sensitivity of our solutions
to these assumptions by examining alternative parameter values below. The realism of the
technological parameters ρ, η, κ will be assessed through simulation under the optimal policy.
Figure 1 gives the regions S, D and I, defined above. In this figure,the x-axis represents
8To see that this boundary condition corresponds to reflection, i.e. to the fact that if Ct hits the boundary,then it is pushed back at infinite speed, note that if we had say ∂
∂C P qt > 0, and it hit the barrier from below,
then a long position in P would be sure to gain in value, which is an arbitrage opportunity.9A relatively high relationship premium may be appropriate for a smaller or younger firm. See Petersen
and Rajan (1998).10Smith(1977) estimates direct underwriting costs for seasoned equity issues to exceed 6% on average rising
to over 13% on smaller issues. He also documents a significant price impact and other indirect costs.11Like us Leland (1998) works with a model where absolute priority is respected in bankruptcy and assumes
proportional bankruptcy costs of 25%. Anderson and Sundaresan (1996) show that observed yield spreadscan be replicated assuming lower bankruptcy costs in a model allowing for strategic debt service.
9
expected revenues ρ, and the y-axis represents the cash holding, which corresponds to short
term debt, at negative values. The ‘save’ region S is denoted by diamonds. As explained,
the solution will not stray beyond the boundary of this region, and the firm does not pay
dividend in this region. The ‘issue’ region is restricted to the lower boundary of cash {C(ρ)}ρ.
We see that the firm is able to borrow short term when ρ greater than about 0.14. Recall
that the firm can only borrow short term, to the extent that all borrowing is riskless.
We also see in Figure 1, that the lowest value of ρ at which the firm will issue shares
if the cash reserve hits the lower boundary, is about 0.08. But the region S bulges to the
left above this value, and is above zero for ρ extending down to about 0.07. At such values
of ρ the firm is incurring operating losses, but there are C values such that (ρ, C) is in the
interior of S. At such (ρ, C) the firm will use the liquidity represented by C to pay operating
losses, in the hope of surviving until business conditions improve. But if (ρt, Ct) strays to
the boundary of S, then the cash reserve will be paid out, and the firm will be abandoned.
We will later discuss covenants preventing such a discrete liquidating dividend.
The downward triangles in Figure 1 represent where the save region S meets the dividend
region D from below. This can be regarded as the liquidity target. When (ρt, Ct) is below
this target, then earnings are retained so as to increase Ct, and when (ρt, Ct) is above this
target dividends are paid immediately, so as to reach this target.
The solid line in Figure 1 shows the average cash holding is a hump-shaped function of
ρ. It is increasing in ρ for low levels of profitability and decreasing for high profitability. It
is far below the target when ρ is below the long-term average, ρ. In this range, the firm is
attempting to increase its cash holdings through retentions, but the dynamics of cash for
given ρ are such that the firm settles down to an equilibrium cash holding short of its target
level. The maximum average cash level of about 0.01 is attained slightly above the long-term
average revenue rate of ρ = 0.15. For revenue rates much above that level the firm is able
to maintain cash at or just below target levels.
The non-monotonicity of cash holding as a function of ρ provides an immediate way of
understanding the finding in the literature where the rate of cash flow has an ambiguous
effect on holdings of cash. Kim et al (1998) find a negative influence of cash flow. Opler
et al (1999) find a positive effect in most specifications. Dittmar et al (2003) and Anderson
(2003) find the sign of the cash flow variable is sensitive to the data set used and the model
specification. In light of our theoretical results, such an unstable pattern is precisely what
we should expect. A similar remark holds for the findings of Opler et al in their Table 3
where they sort firms in quartiles based on cash holdings and see that cash flow rates first
rise then fall with average cash holdings.
Other properties of the model can be seen in Figure 2 where we present a portion of one
10
history of the model when simulated under the optimal policy for the benchmark parameter
values. This depicts the last ten years of the firm ending with its bankruptcy in its 127’th
year. After the decline in profitability in year 119 the firm has exhausted its cash reserves and
survives by issuing equity. With the recovery of revenues in year 121 above about ρ = 0.145
(compare with Figure 1) the firm briefly takes on some short-term debt; however, the amount
of such debt is so small as to be imperceptible on the chart. With the subsequent strong
recovery of revenues, the firm begins to accumulate cash until attaining the cash target of
something less then 0.02 at which point it begins to pay dividends. (Again, compare with
Figure 1). With the sustained, sharp fall of revenues from year 125 though mid-year 127,
the cash reserve is exhausted, the value of equity drops to zero, and the firm is bankrupted.
2.5 Benchmark Simulation of the Model
One of the strengths of the contingent claims framework is that the model can be used to
derive a great many implications for observable firm characteristics which can be used in
calibrating and testing. To explore our formulation of the model we simulate it as in Figure
2 and summarize its implications for the liquid reserves, leverage ratios, equity volatility,
credit spreads, default probabilities, and other relevant measures.
Specifically, for the parameters above, we solve the model for the equity and debt values
as functions of (ρ, C), and obtain the optimal regions D, I and S. Then we perform 300
simulations of the ρt variable, the realized cash holding Ct, equity value Jt and debt value
Pt. Each simulation starts from ρ = 0.2, C = 0.0 and runs to the firm’s bankruptcy, or to
1000 years, if no bankruptcy occurs before then. Based on the realizations of the simulations
we also calculate the average liquidity for each grid value of ρ, denoted C(ρ). Taking this
average liquidity value as a function of ρ, we study the model conditional on 4 levels of
profitability: ρ = 0.10 (‘low’), ρ = 0.15 (‘normal’), ρ = 0.20 (‘high’), and ρ = 0.25 (‘very
high’). At each level of ρ and average liquidity C(ρ), we present the net equity12 value
Jq(ρ, C(ρ)) − C(ρ) (i.e. equity, net of the liquid reserve), debt value P q(ρ, C(ρ)), net firm
value Jq(ρ, C(ρ))+P q(ρ, C(ρ))−C(ρ) , leverage (debt value divided by the value of the firm),
and the yield spread (yield on debt less r). Finally, we give the equity volatility, calculated
as
√(√ρη ∂
∂ρJ
)2+
(σ ∂
∂CJ
)2/J .
These simulations are done with respect to the objectively realized (“statistical”) prob-
abilities, and so we need to specify a risk premium. Referring to the risk associated with
ρ, we can represent the risk premium by a parameter λ (assumed constant, for simplic-
12Taking net equity and firm values facilitates comparisons across different scenarios, since changing be-tween scenarios would involve making up the difference in the liquidity reserve by cash.
11
ity)13, such that to transform from the risk neutral to the statistical measure, we should
replace dW ρt of Equation (2) by dW ρ
t + λdt. This λ can be thought of as a Sharpe Ra-
tio: it is the extra return required, per unit of extra exposure to the risk represented by
dW ρt . To see how the risk premium affects the return of the firm’s equity, note first that
under the risk neutral measure, the expected return will be just the riskless return r. The
equity value Jt(ρ, C) is a smooth function of ρ, and using the Ito formula, we can write
dJt = (drift)dt + ∂∂ρ
Jtdρ ≡ rdt +√
ρtη∂∂ρ
JtdW ρt . The factor
√ρtη
∂∂ρ
Jt here is the equity
volatility, and we will calculate this in the tables below. Substituting dW ρt by dW ρ
t + λdt,
we can see that the risk premium increases the expected return by λ times this volatility.
A reasonable value for the risk premium (Sharpe Ratio) of the market itself in λ = 0.5,
corresponding to a market excess return of say 8%, and market volatility of say 16%. On
the other hand a completely diversifiable risk would imply λ = 0. We take λ = 0.3, which
is reasonable, if we assume that the risk of the firm has a systematic component, i.e. it is
somewhat correlated with the market.
In addition we calculate the yield spread on zero-coupon bonds of 5 and 20 years until
maturity. For this calculation, and following Duffie and Lando (2001), we assume that the
perpetual debt is made up of a continuum of zero coupon bonds, and if the firm defaults,
the these bonds are paid off in proportion to their value weight in the total debt. This
calculation is done by adapting the perpetual debt valuation to accommodate this default
rule, a payment of one dollar if there is no default before maturity, and the coupon being
zero. We also calculate the probability of bankruptcy at 1, 5 and 20 year horizons. This
calculation is again done by adapting the perpetual bond valuation, and we include the risk
premium λ, since this probability is not risk neutral, but objectively realized.
Our results for liquidity, debt value, equity and leverage values are given in Table IA,
and the credit relevant values are given in Table IB. These tables also contain results for
other values of the debt parameter, which we will discuss later. We take as the main ref-
erence for our calibration the case q = 0.004 and ρ = 0.15 which corresponds to normal
business conditions. In Table II we have summarized the results for this case in a way that
can be compared to financial ratios of US non-financial firms as reported by Standard and
Poors for the period 1997-1999. These comparisons suggest that our benchmark firm has
characteristics similar to a firm toward the bottom end of the investment-grade range, BBB.
Also from Table IA, we see that in this case, the average liquid asset of about 10.5% of
total asset value which is in line with average liquidity holding for small to medium sized
firms during the 1980’s and 1990’s as depicted in Opler et.al Figure 2. Opler et.al. also
investigated the dynamic properties of cash holding by U.S. firms and found evidence of
13Such λ has to exist, in the absence of arbitrage: see Duffie (2001).
12
mean reversion suggesting that firm target cash holdings. We have compared our model to
these results by estimating their Equation 1 using simulated data of our model under the
optimal policy using the benchmark parameters. The results are presented in Figure 3 and
are quite similar to those reported by Opler et.al. in their own Figure 3.
Our model implies an equity volatility of 40% under normal conditions (ρ = 0.15) which
is in line with market experience. For example, Zhang et.al. report that for a sample in
which BBB rated firms predominate, historical average daily equity volatility ranges from
40% to 50% per year.
Turning to the credit risk related indicators in Table IB, we see that this firm in the
reference case has a credit spread of 90 basis points on the 5-year pure discount bond. By
way of comparison, Zhang et.al. report that the average 5-year CDS (credit default swap)
rate on names rated BBB was 116 basis points between 2001 and 2003. During the same
time the average 5- year CDS rates on firms rated A or above were 45 b.p.’s; whereas, the
comparable rates on high yield CDS’s were 450 b.p.’s. These may be compared to the spreads
implied by our model for the cases ρ = 0.2 (59 b.p’s) and ρ = 0.10 (397 b.p.’s), respectively.
From Table IA we also see that, under our assumption about the risk premium, the
probability of default at the five-year horizon is 1.8%. This is corresponds to Standard and
Poor’s historical experience for a bond rated BBB or perhaps a bit below (see, de Servigny
and Renault).
Finally, our model has implications of the value of defaulted bonds which can be compared
to empirical recovery rates. Specifically, Figure 4 gives the equity valuations in terms of ρ, at
C = 0 for our benchmark case (circles). The corresponding value of collateral is calculated
as (1−α) times the value of unlevered equity and is depicted by the solid line. We see from
Figure 1 that the benchmark firm will go bankrupt for ρ in the neighborhood of ρ = 0.08
depending upon the path of liquidity. Thus from Figure 4 we see that the value of debt of
the bankrupted firm is about 0.03. From the same figure we also see that the debt value
is about 0.06 for ρ ≥ 0.15. Assuming that the long-term debt was issued at par with
business conditions in this range, we see that the debt recovery rate on bankruptcy is about0.030.06
= 50%. This number is consistent with empirical studies of BBB rated firms. Over the
period 1988-2002 Standard and Poors found that the recovery rates on defaulted bonds were
30 per cent for Senior Subordinated Notes, 38 per cent for Senior Unsecured Notes, and 50
per cent for Senior Secured Notes. (See de Servigny and Renault (2004) Chapter 4.)
To summarize, by simulating our model under the optimal policy and for the benchmark
parameters we see it is consistent with a very wide range of empirical benchmarks. These
include cash holdings, leverage, equity volatility, yield spreads, default probabilities and
recovery rates.
13
3 Some Applications of the Model
3.1 The Dynamics of Leverage and “Optimal” Capital Structure
It has long been recognized that the theory of optimal capital structure based on a static
trade-off of tax benefits, bankruptcy costs and other financial frictions is difficult to square
with a variety of empirical regularities including mean reversion of firm leverage ratios and the
negative correlation of leverage ratio and lagged profitability. The initial work on dynamic
trade-off models by Fischer et.al. suggested that some of the observed inertia in capital
structure might be explained by fixed adjustment costs. More recent efforts with dynamic
trade-off models, notably by Hennessy and Whited and Strebulaev, have shown that under
plausible values of parameters these models are capable generating mean reversion of leverage
and its negative correlation with lagged profits.
Our model may be viewed as a further development in research on dynamic trade-off
theory. Like Hennessy and Whited we allow for both the accumulation of cash and for
variable rates of short-term borrowing. However, unlike their study, we work in a continuous
time contingent claims framework and we allow for both short-term and long-term debt. Our
model is also capable of generating mean reversion of leverage and the negative dependence
of leverage on lagged earnings. In Figure 5 we present simulation results for our model
with a constant level of long-term debt under the optimal policy and with our benchmark
parameters. Leverage, measured by the book value of total long-term and short-term debt
divided by the firm value, tends to fluctuate around the mean value of 0.44. The regression
coefficient of -2.6 in Figure 5 corresponds to a significant correlation of -0.47 between leverage
and expected revenues (ρ) lagged one year.
In our model this behavior is the by-product of the share-value maximizing policy toward
dividends and short-term borrowing. As seen in Figure 1, over a range of revenues close to
their long-term mean an improvement in revenues (and earnings) will result in higher levels
of cash-holdings which reduces the book value of total debt while firm value is increasing.
The adjustment of cash to earnings is not one-for-one for as earnings improve the targeted
cash level is reduced and eventually the higher earnings will be paid out.
Since our model includes both long-term and short-term debt we are able to explore the
trade-off between these two and also their interaction with cash holding and dividends. The
key insights can be understood from Figure 6 which depicts the optimal policies for two levels
of long-term debt. The “high debt” case corresponds to our benchmark where q = 0.004 and
is depicted by lines drawn with ‘x’. As in Figure 1 we present the target level of liquidity,
the average liquidity, and the debt capacity and equity issuance boundary as functions of
expected revenues, ρ. In the “low debt” case we set q = 0.0 with all other parameters as in
14
the benchmark case. The solutions are depicted with circles in Figure 6.
We see from Figure 6 that the effect of higher levels of long-term debt is to raise the
liquidity target for any given value of expected earnings. Similarly, at a given value of ρ, the
equity issuance boundary is increased (i.e., debt capacity decreased) and the average level
of liquidity is increased. It will be noted that in the low debt case the average level of cash
holdings is negative for all ρ. That is, on average the firm engages in varying degrees of
short-term borrowing as a function of business conditions.
It is clear from this analysis that long-term and short-term debt are highly substitutable.
With a reduction in long-term, a given firm will compensate by reducing its cash holdings
and, possibly, borrowing short-term. In this way it will achieve a similar balance of debt tax
shields and bankruptcy costs. It will be noted from Figure 6 that for values of ρ ≥ 0.19 the
average cash and target cash levels coincide for both high-debt and low-debt firms, implying
that the dividend behavior is very similar for the two firms.
What are the implications of this for the optimal “time zero” level of long-term debt as
in static studies of capital structure? For a firm with our benchmark parameters the answer
is given in Table I. For example, if ρ = 0.2 and the firm sets its value of q once for all it
will maximize the firm value by setting q = 0.004, which coincides with the case we have
taken as our benchmark. Thus our model does imply an optimal capital structure in the
traditional sense. However, what is more interesting in Table I is that large variations in the
level of long-term debt have only a small impact on firm value. For example, at ρ = 0.2,
setting q = 0.002, i.e., one half of the “optimal” level leads to a reduction of firm value by
only 0.3% (0.1517 versus 0.1519).
Thus we see that there is a near irrelevance of long-term capital structure in our model.
Whatever the firm sets as its long-term debt level, it can find a corresponding cash and
short-term debt policy that balances off bankruptcy costs, tax shields and other costs so
as to achieve approximately the same value of the firm. Through experimenting with a
number of variations on our benchmark formulation we consistently obtain the same near
irrelevance result. For example, as will be discussed in detail below, if we change the rate
of mean reversion of our model we change some aspects of the model in important ways. In
particular, with a low rate of mean reversion the firm optimally holds a higher level of total
debt, in line with the findings of Leland (1994) who assumes Geometric Brownian Motion.
Nevertheless, in this case as well, the value of the firm is quite insensitive to the level of
long-term debt chosen.14
14Taking κ = 0.4 and setting other parameters as in our benchmark one finds that conditional upon ρ = 0.2the net firm value is maximized at q = 0.007 as compared to q = 0.004 in our benchmark case. However withthese parameters the net firm value is quite insensitive to variations of q. For example, q = 0.004 results in
15
This seems to us one of the deep implications of a dynamic analysis of capital structure
that takes also into account cash holdings and dividend policy, and this would seem to not
depend importantly upon the specific technology or parameters we have adopted. And like
the classic irrelevance results of Modigliani and Miller, it gives a natural way of interpreting
one of the most troubling empirical findings for any theory of capital structure, namely
that for otherwise very similar firms, we can observe a very wide variety of policies on debt
structure. Long-term debt policy may be set according to some rule of thumb that appears
reasonable to management. Once that decision is made, so long as the dividend/retention
policy is adjusted optimally, there is no value maximizing imperative for deviating from the
chosen rule for long term indebtedness.
While our dynamic trade-off model makes no explicit allowance for information asym-
metries it is interesting to compare the model’s implications to the recent re-examination
of the pecking order theory by Leary and Roberts (2006) who attempt to overcome the
problem of low power which characterizes earlier tests of the pecking order.15 They develop
an empirical model of the pecking order that takes into account the state dependent nature
of the pecking order. Depending upon the state, the firm will select financing according to
three alternative hierarchies: (a. pecking order) internal first, then debt, then equity; (b.
debt capacity constrained) cash first, then equity; and (c. cash constrained) debt first, then
equity. These cases are set out in their Figure 1. These three cases also emerge in our model
where the relevant state variables are ρ and C as can be seen in our own Figure 1. A classic
pecking order (case a) prevails in our model for relatively high values of expected revenue
(ρ) where the equity issuance boundary corresponds to negative C. That is, as financing
needs would increase (e.g., as a result of a larger negative cash flow shock) they would be
met first by drawing down internal funds, then by borrowing short-term and then by issuing
equity. Case b corresponds to lower values of ρ where the equity issuance boundary coin-
cides with C = 0. And the cash constrained case (c) would correspond to high values of ρ
where current cash is negative. Thus our dynamic trade-off model provides an alternative
rationale for their empirical model which does not rely on positing asymmetric information.
It is notable that Leary and Roberts find no evidence that the predictive performance of the
empirical pecking order is related to indicators of information asymmetry. However, they
do find that predictive performance is improved by allowing the parameters of the model
to depend upon the covariates such as cash flow volatility and a dummy variable indicating
that the firm pays dividends which are entirely in line with our model.
a net firm value about 1% less than for the q = 0.007.15Most earlier tests of pecking order followed the approach introduced by Shyam-Sunder and Myers (1999).
16
3.2 Asset Substitution Effect and Hedging
Since the analysis of Jensen and Meckling focussed attention on the incentive for shareholders
of the levered firm to engage in ex post increases in asset volatility, the “asset substitution
problem” has become one of the pillars of received wisdom in corporate finance. Based
on this, there is a large literature on how this problem can be remedied either by security
design (with Green (1984) being an early contribution) or by choice of capital structure (with
Leland (1998) giving an explicit analysis of the problem). It is important to realize that the
intuition about the asset substitution effect has been derived within largely static models,
and it remains to be seen whether this is also a feature within a dynamic setting such as
ours.
In our model, the parameters η and σ represent different sources of earnings volatility.
η refers to the dynamic of the profitability ρ. Given the mean reverting nature of the
profitability relation, η shocks are persistent. In contrast, σ refers to a white noise type of
volatility which represents a non-persistent shock to profitability.
The effect of varying these parameters are seen in Table III where we have calculated the
sensitivities of variables of interest to the main economic parameters of our model. As in
Table I, each column is based on simulations of our model under the parameters specified and
under the corresponding optimal policy. Values of the dependent variables are calculated
conditional on ρ and the associated average value of C in the simulation.
In Columns 2 and 3, we examine the effect of varying the level of the persistent volatility
while the other parameters are at their benchmark levels. These columns should be compared
with each other, and with Column 1, which repeats the results for the benchmark case.
Focussing on the case of ρ = 0.15, we see that the value of debt and the value of equity are
both decreasing in η. As judged by the percentage decrease in value it appears that debt
is more sensitive than equity to increases in η, but the direction of the effect is the same.
A similar pattern holds for high profitability business conditions, ρ = 0.20. Interestingly, it
holds as well when the firm is not far from bankruptcy, ρ = 0.10.
Thus in contrast with standard learning on the asset substitution effect, in our dynamic
setting both debt and equity seem to have similar incentives to avoid increases in persistent
cash flow volatility. Why is this the case?16 Note that for each ρ, the level of cash holdings
is increasing in η. Thus in response to an increase in volatility, under the optimal cash and
dividend policy, the shareholders would choose to increase the amount of cash that they
hold. This increased use of internal slack comes at a cost, and equity values are reduced as
16De Marzo and Sannikov who also study a continuous time model with cash accumulation and short-termand long-term debt also find this convergence of interests of debt and equity. They characterize the resultas “surprising” but do not pursue the issue further.
17
a result.
The effects of changes in nonpersistent volatility, σ, are seen in columns 4 and 5 of Table
III. We see that for each level of ρ increasing σ again gives rise to lower values of both debt
and equity. The reason that shareholders are hurt as well as are creditors is that under the
optimal policy shareholders hold a higher liquid reserve.
Thus for both persistent volatility and transient volatility, the interests of debt and equity
are roughly aligned for the changes of the volatility parameters that we have considered. This
does not mean that they are exactly aligned and that they would agree upon exactly the
optimal value if that parameter alone were to be varied.17 However, it does raise a question
about the empirical relevance of standard teaching on the asset substitution effect. Indeed,
empirical studies have been notably unsuccessful to find any evidence in support of asset
substitution effect (see, Andrade and Kaplan 1998 and Rauh 2006).
These results also relate to the literature on corporate incentives to hedge that has grown
up on recent years. For example, in continuous time frameworks Mello and Parsons (2000)
and Rochet and Villeneuve (2004) have characterized the benefits of hedging non-persistent
shocks to cash flow. While our framework and our major focus is different than theirs, our
results are similar to theirs if we assume that σdW σt represents a shock to earnings that can
be hedged costlessly (resulting in σ = 0). For example, if the source of transient shocks
to earnings come from fluctuations in commodity prices it may be possible to eliminate
these risks with positions in short-dated futures contract.18 As in our discussion of the asset
substitution effect, it is clear that benefit of hedging to shareholders comes from the fact
that it would allow the firm to reduce its use of costly cash balances.
Our results on increases of η suggest that shareholders would also have an incentive to
hedge persistent fluctuations of revenues (ρ). However, it is not clear that suitable hedging
instruments are available or that the costs of hedging this risk would be low.19.
Finally we note that our results confirm one of the most basic findings of empirical studies
of liquidity, namely that higher volatility is associated with higher liquid asset holding in
cross sections. (e.g., Opler et al).
17Stated otherwise, for some parameter values and in some states it may be the case that equity will beincreasing and debt value will be decreasing in η.
18Note that σdW σt is a martingale difference, corresponding to the assumption that there is no risk premium
associated with this risk. An imperfect hedge may result in a reduction of σ.19A costless hedge of ρ would not be modelled by simply taking η = 0, but rather by subtracting the
martingale component of the process ρ.
18
3.3 Mean Reversion
Our assumption of mean reversion of the firm’s revenue process stands in contrast with
the assumption of Geometric Brownian Motion adopted in much of the earlier literature on
contingent claims analysis which emerged as a direct outgrowth of the contribution of Black
and Scholes.20 Mean reversion is a simple way to capture business cycle effects and also the
dynamics of innovation and imitation in competitive product markets. Similar assumptions
about the firm’s technology have been made by Gomes (2001) and Hennessy and Whited
(2005) who also provide supporting empirical evidence.
The effects of changing the speed of mean reversion can be seen in Table III, columns 6
and 7 where we have varied κ. Increasing the speed of mean reversion tends to reduce the
influence of shocks to expected revenues and thus might be seen as something like a decrease
in η. In fact the effects of the two parameters are somewhat different. This can be seen
with reference to the case ρ = 0.15 where debt and net firm values are both increasing in
κ; whereas, net equity is decreasing. The divergence of the interests of debt and equity are
stronger under good business conditions when equity would be aided by a decrease in the
speed at which the firm returns toward the long-run mean. This is seen in the case ρ = 0.2
where the value of equity is strongly decreasing in κ; whereas debt is increasing. However,
when the firm is close to bankruptcy, ρ = 0.1, both equity and debt would benefit from
faster mean reversion. To summarize, faster mean reversion is good for shareholders when
business conditions are poor and bad for shareholders when business conditions are good.
In contrast, faster mean reversion benefits creditors in all business conditions.
One criticism that has been leveled at static trade-off models is that they predict unreal-
istically high optimal leverage. For example, this was the case in Leland (1994), a fact that
motivated the study of refinancing boundaries in Leland (1998). In this formulation when
firm value has risen sufficiently relative to the point at which the current debt structure
was fixed, shareholders will increase the amount of debt issued by a discrete amount. When
such financial restructuring is possible, the optimal “time zero” degrees of leverage are more
in line with observed leverage ratios. Strebulaev (2007) pursues this line of reasoning and
shows that potential restructuring can help to explain a number of other empirical puzzles
related to leverage.
The incentive to restructure long-term debt under good business conditions will be re-
duced by the presence of mean reversion. For example, even under the assumption of very
low restructuring costs (1%), Leland found that refinancing is implemented after the asset
20Recent contributions that make the assumption of Geometric Brownian Motion include Leland (1994)and (1998), Mello and Parsons (2000) Rochet and Villeneuve (2004).
19
value has increased to about twice its initial value. In our formulation, firm value is very
unlikely to increase by this amount. For example, in our benchmark simulations that were
initiated at ρ = 0.2, revenues never increases beyond ρ = 0.3. From Figure 4 we see that
ρ = 0.3 corresponds to a value increase of about 1.5 fold. In fact, from such a very high level
of expected revenues, the firm’s profitability will then be predicted to fall rapidly through
the force of mean reversion.21
3.4 Other Comparative Statics
The remaining columns in Table III pertain to variations in what might be thought of as
“efficiency” parameters describing the environment where the firm operates.
In Columns 8 and 9 of Table III, we vary the proportional bankruptcy cost α. In our
benchmark we have set this at 30% of the value of assets in place at the time of bankruptcy.
This may seem high compared to some studies in the literature, e.g., Warner (1977). How-
ever, those estimates pertain to direct bankruptcy costs typically for firms with large amounts
of tangible assets and costs are expressed as a proportion of the book value of assets reflecting
historical costs. Furthermore, more recent studies covering indirect costs and a wide range of
industries (including those with substantial intangible assets) suggests that total bankruptcy
costs may be very substantial. (See Franks and Torous (1989) and Weiss (1990)). By way of
comparison, Leland (1994) assumes 50% proportional bankruptcy costs and Leland (1998)
assumes 25%.
In Table III we consider a low (α = 0.05) and high (α = 0.5) bankruptcy cost scenarios to
compare with our benchmark. As expected the effect of higher bankruptcy costs is to depress
the value of long-term debt. We find that higher bankruptcy costs also hurt equity values.
This may seem surprising in our model where absolute priority is enforced in bankruptcy.
The reason for this effect is that variations in bankruptcy costs affect the firm’s capacity to
borrow short-term and therefore has an impact on the optimal cash holding policy of the
firm. In general, higher bankruptcy costs induce higher levels of cash holding which tend to
depress equity values.
While this link between bankruptcy costs and equity is rather less direct than its impact
on debt values, the effect can be substantial. For example, with ρ = 0.15, in the case of
very low bankruptcy costs = 0.05, the firm’s average cash holdings are 25% less than in the
21When mean reversion is weak in our model we approximate the results found under the assumption ofgeometric Brownian. For example, as has already been pointed out, with κ = 0.4 the static optimal levelof long-term debt is much higher than in our benchmark (q = 0.007 versus q = 0.004). Also, with κ = 0.4,the likelihood that ρ will persist at levels far above the long-term mean increases. Accordingly, there will begreater incentives to engage in costly adjustments of financial structure a la Leland (1998).
20
benchmark (0.0104 versus 0.0136) and equity values are 2.8% higher as a result (as compared
to debt values which are 3.5% higher than in the benchmark).
In our model the efficiency of external capital markets is captured by the parameter θ
which is the fraction of the value of securities issued by the firm which accrues to the firm.
The fraction 1− θ is lost through the underwriting process as commissions, fees or excessive
dilution. As we have already noted our benchmark assumption of θ = 0.8.
In Column 10 of Table III, we consider the extreme case of θ = 0.01, which may be
taken as the case when capital markets very inefficient. In Column 11 we take the opposite
extreme of θ = 0.95 which would seem to correspond to a highly efficient capital market and
small agency costs. While θ cannot be directly interpreted as the degree of underpricing, the
estimates of the underpricing of IPO’s, range from 5 per cent to often greater than 20 per
cent. In the U.S. seasoned issues typically involve somewhat lower costs. From the estimates
of Lee et.al. (1996) using the U.S. data, reasonable estimates might be 11% for IPO’s and
7% for seasoned issues. This suggests that θ = 0.95 would be appropriate only for very
highly developed capital markets, and where there is no information asymmetry.22.
From Columns 10 and 11 of Table III we see that when capital market efficiency increases,
the average level of liquid asset holding decreases. Focusing on the average profit case ρ =
0.15, liquid assets average about 50 per cent of the net firm value in the very underdeveloped
capital market context. In the 95 per cent efficient capital market, optimal liquid assets
correspond to only about 3 per cent of firm value. This result is consistent with Opler et
al (1999) who document empirically that firms with greater access to capital markets carry
smaller amounts of liquidity reserve. It is also in line with the a study of listed firms in
twelve continental European countries by Ferreira and Vilela (2002). They find a negative
association between liquidity and capital market development, proxied by the ratio of the
country’s free float (i.e., total value of stocks held by minority shareholders) to GDP.
In our framework the cost of holding liquidity inside the firm is represented by the r−rin.
This is a proxy for the amount of rent extraction that is obtained through various forms of
managerial moral hazard. In Columns 12 and 13 of Table III we take rin = 3% and 5%
while we maintain our other benchmark parameters including r = 6%. In our view rin = 3%
would indicate severe agency problems since in this case half the income flow from liquid
assets would be diverted or wasted by managers.
As we would expect, the average level of liquid asset holding is increasing in rin. The
effect is quite strong. Focussing on the normal profit case (ρ = 0.15), when rin = 0.05 the
optimal level of liquid asset holding is about 16 per cent of net firm value as compared to
22Based on a sample of US, non-financial firms between 1993 and 2001 a period of very high stock marketactivity Hennessy and Whited (2005) estimate floatation costs of about 5.9%
21
12 per cent in the benchmark case. The values of both equity and debt are increasing in
rin. The benefit to shareholders is direct since as residual claims on the firm’s cash flows any
increase in the return on a given level of liquid securities accrues to them. The benefit to
creditors is indirect through the fact that the higher level of cash holding by the firm reduces
the chances of bankruptcy. Most of the benefit of lower costs of holding liquid assets accrue
to shareholders. For example, when compared to the benchmark, the value of equity is 2.2
per cent higher when rin = 0.05 in contrast with long-term debt whose value is increased by
half a per cent.
The last two columns of Table III show the effects of varying rbank. The difference rbank−r
is a proxy for the informational rents ceded to creditors through short-term borrowing. An
increase in rbank discourages short-term borrowing and induces the firm to carry somewhat
greater amounts of cash on average. This is associated with lower values for shareholders
and higher values for long-term creditors. However, these value effects are slight and only
noticeable under good business conditions ρ = 0.2.
Finally, we note an interesting indirect effect that applies to the efficiency parameters θ,
rin and rbank. Holding q and other parameters constant, a change in the efficiency parameter
that induces greater liquid asset holding will be associated with lower equity volatility. So,
for example, a change in stock market regulation that promotes efficiency in equity issuance
(i.e., higher θ) may indirectly induce greater stock market volatility.
4 Extensions
4.1 Growth Opportunities
4.1.1 Background and empirical context
Our basic framework is rich in modelling a wide range of capital market operations available
to the firm. However, it is restrictive in that the firm’s risky asset is fixed. This simplification
is useful because it gives rise to a stationary problem from which we have been able to derive
a number of clear and strong implications. However, the simplification comes at the cost of
not allowing for firm growth. The presence of growth opportunities is likely to increase the
incentive to retain earnings. In this section, by extending the model to allow for growth,
we examine the robustness of the qualitative conclusions derived above as well as measure
quantitatively the sensitivity of optimal cash holding to changes in the parameters affecting
growth.
To specify more precisely the nature of the growth opportunity, it will be useful to
22
examine some of the main empirical facts about growth and related major issues of long-
term financial contracts. In Table IV we summarize average annual capital expenditures,
share issuance and long-term debt issuance as a fraction of preceding year total assets for
US non-financial firms between 1978 and 2008. We report these averages cross-tabulated by
profitability quintile and leverage level. The data are from Compustat.
The first panel shows that capital expenditures tend to be highest for the firms with
highest current profitability. This holds overall and for each leverage category. Overall the
amount of growth of capital stock is moderate. Assuming a depreciation rate of 6% per year
(see e.g., Nadiri and Prucha, 1996), firms in the highest quintile expand their fixed capital by
about 4% on average. This is consistent with the notion that organic growth which increases
the scale of the firm’s existing line of business is the dominant mode of firm growth. Growth
is not a monotone function of profitability. The lowest quintile by profitability tend to have
greater capital expenditures relative to existing assets than firms in the second quintile of
profitability. This may well reflect the influence of younger firms just building up their
productive capacity.
From the second panel we see that share issuance tends to be highest for the least prof-
itable firms. This is consistent with the prediction of our baseline model whereby seasoned
share issues occur optimally in times of relatively low profitability (see Figure 1). Interest-
ingly, share issuance is negative among the highest quintile by profitability both overall and
in all but one of the leverage categories. That is, share buy backs are widely used to payout
earnings when economic conditions are exceptionally good. From the third panel we see it
is also the case that long-term debt issuance is lowest for the highest earning firms. Taken
together these results suggest that a dominant form of firm growth occurs among relatively
profitable firms and that to a significant extent this growth is financed through internal
resources rather than new issues of equity or long-term debt.
4.1.2 Modelling growth
A model of investment capturing the features of organic growth just described can be de-
veloped as follows. We analyze the firm with a one time opportunity at time T to increase
productive capacity, up to a certain limit. The investment results in the rescaling of the
firm’s existing asset in place at the proportion (V + I)/V where I is the expenditure on in-
vestment and V is capacity prior to undertaking growth at time T . We assume I ≥ 0. Note
that since investment is measured in monetary terms, the cost of new productive capacity,
i.e., ‘machines’, is given implicitly by V . That is, a low value of V implies that new machines
are cheap and a given expenditure on investment will result in a large increase in the scale
of the fixed asset.
23
The new investment can be financed through some combination of equity issue, long-term
debt issue, and drawing down cash (or short-term borrowing). For tractibility, we assume
that long-term debt issue takes the form of a rescaling of contractual coupon payments q
in proportion to investment. The firm chooses the scale of investment, the amount of cash
financing and the amount of equity financing to maximize the value of equity. As in section
2.2 we denote the value of equity without growth option as Jq(ρ, C). As mentioned above,
this firm has scale V . Extending this notation, we will write the equity value after growth
of physical assets and coupon to scale V when cash holding is C as Jq(ρ, C, V ).
Since we have assumed constant returns to scale, proportionate taxes, equity issuance
costs, fixed costs and bankruptcy costs, the value of equity will be homogenous of degree 1
in the cash flow parameters, coupon and cash. That is, Jq(ρ, kC, kV ) = kJq(ρ, C, V ). By
the properties of linear homogenous functions we can show,
Jq(ρ, C, V ) = VV
Jq(ρ, C VV
, V ), (10)∂
∂CJq(ρ, C, V ) = ∂
∂CJq(ρ, C V
V, V ), (11)
∂∂V
Jq(ρ, C, V ) = (1/V )Jq(ρ, C, V )− (C/V ) ∂∂C
J(ρ, C, V ) (12)
These equations (and similar equations for P in place of J) enable us to calculate the values
and derivatives of Jq at an arbitrary scale V , from their values at scale V , where the latter
can be calculated numerically, using the procedures of Section 2. This will prove useful in
the analysis below. Note that growth by increasing the scale of fixed assets and coupon while
keeping cash constant will exhibit decreasing returns to scale.23
We proceed recursively by first analyzing the growth decision at the instant T of the
growth option. Then we will study optimal cash holding and dividend policy prior to growth
when this opportunity is merely a prospect which might be stochastic. Specifically, at each
grid point (ρ, C) we calculate the optimal action of the firm, in terms of how much to grow,
how much cash to draw down, and how much equity to issue. This gives us the equity value
Jq,GRW (ρ, C) at the instant before growth. Second, we analyze the cash reserve policy that
the firm will pursue before time T , trading off the inefficiency of holding cash against the
inefficiency having to issue equity in order to exploit the growth opportunity. This second
step simply involves developing the solution back from Jq,GRW (ρ, C) at time T , using the
same procedure as in Section 2.
We now deal with the first step above, and describe our procedure for finding the optimal
growth decision at time T , from scale V , and at any point (ρ, C), and the corresponding
23The problem of irreversible investment with decreasing returns to scale was analysed in a certaintysetting by Arrow (1968) using variational methods. Our formulation adapts this analysis to a stochasticgrowth setting using our computational framework.
24
equity value Jq,GRW (ρ, C). First, suppose the firm’s current scale is V ≥ V , and it is
considering to increase its scale by a small amount dV . For the time being, assume that the
firm will not issue equity, but will finance the growth by drawing down cash in the amount
dC, and raising the rest, denoted dP , from issuing long term debt. After growth to scale
V + dV , the debt value will be P q(ρ, C − dC, V + dV ), and the firm will have raised an
amount P q(ρ, C − dC, V + dV ) dVV +dV
from the issue. Thus
dV = dC + dP ≡ dC + P q(ρ, C − dC, V + dV )dV
V + dV, (13)
and letting dV → 0 givesdC
dV= 1− 1
VP q(ρ, C, V ). (14)
This equation tells us how much cash must be drawn down, if the firm grows from scale V
to scale V + dV , with no equity issue, but a proportionate increase in long term debt.
If the firm decides to grow by dV by drawing down dC from the cash reserve, then the
gain to equity will be
Jq(ρ, C − dC, V + dV )− Jq(ρ, C, V ) =
[∂
∂VJq(ρ, C, V )− dC
dV
∂
∂CJq(ρ, C, V )
]dV + o(dV )
(15)
The firm will thus find it profitable to grow by issuing long-term debt in step with the growth
of fixed assets and by drawing down cash as required to cover the costs of investment if
∂
∂VJq(ρ, C, V )− dC
dV
∂
∂CJq(ρ, C, V ) > 0 (16)
where dCdV
is as in Equation (14).
Now instead suppose the firm decides to grow dV by issuing debt as above and equity by
an amount dQ sufficient to cover the costs of investment while keeping cash constant. Then
the amount raised by debt will be P q(ρ, C, V + dV ) dVV +dV
, and corresponding to Equation
(13), we have
dV = dQ + dP ≡ dQ + P q(ρ, C, V + dV )dV
V + dV, (17)
and letting dV → 0, this gives
dQ
dV= 1− 1
VP q(ρ, C, V ), (18)
whose RHS is the same as that of Eqn (14). The firm will thus find it profitable to grow
by issuing long-term debt in step with the growth of fixed assets and by issuing shares as
required to cover the costs of investment if
25
∂
∂VJq(ρ, C, V )− dQ
dV
1
θ> 0. (19)
In fact, the firm can choose to grow by issuing equity or by drawing down cash. It will
choose the former if (16) holds, and LHS(16) > LHS(19); and it will choose the latter if (19)
holds, and LHS(19) > LHS(16). Actually, our valuation of Jq(ρ, C) is Section 2 includes the
assumption that the firm will increase the cash reserve from C by issuing equity, if this is
profitable, which corresponds to the criterion ∂∂C
Jq(ρ, C) > 1θ. This, and Equation (11) give
∂
∂CJq(ρ, C, V ) ≤ 1
θ, (20)
and if we have equality in Equation (20), then the firm should issue equity to increase the
cash reserve.
Our procedure for determining the growth decision at time T at each grid point (ρ, C)
is as follows. Let Vmax be the maximum possible scale. We take Jq(ρ, C, V ) as calculated
in Section 2. Starting from the initial scale V , we increase the scale in small steps of size
dV , i.e. to Vi := V + idV , for i = 0, 2, 3, ..., and decreasing C by steps dCi
dVdV , i.e so that
Ci = Ci−1 − dCi−1
dVdV for i = 1, 2, 3, .... We continue until the highest i , say i = n, such that
(16) holds or until Vi = Vmax. The value Jq,GRW (ρ, C, V ) immediately before growth, is then
the value Jq(ρ, Cn, Vn)
At this point it may be that cash has been depleted to the point that would make equity
issuance preferable. Therefore, if at (Cn, Vn) calculated above, we have ∂∂C
Jq(ρ, Ci, Vi) = 1θ,
then we take the firm to issue equity to replenish cash. This is done in successive increments
dQ to Qi = i×dQ, i = 1, 2, 3, ... resulting in Cn,i = Cn+i×dQ, until we have ∂∂C
Jq(ρ, Ci, Vi) <1θ.
We finally discuss the case when the growth opportunity is stochastic, in particular, when
the firm is uncertain whether the opportunity with actually occur, but if it does, then the
timing and size are known. In this case, the value the equity at each (ρ, C) and at the growth
time T , is the probability weighted average of the value assuming that the growth does occur
(calculated as above), and the value assuming that the growth does not occur (which is just
the value in Section 2). This calculation assumes that the growth probability is taken as risk
neutral. Also, we assume that the firm knows whether the growth opportunity has occurred,
when it implements its cash draw down, equity and debt issuance, at time T .
This completes the analysis of the optimal growth and financing decision for each node
(ρ, C) at time T . Given the resulting value of equity,Jq,GRW (ρ, C, V ), we then proceed to
solve for the optimal dividend and share issuance policy in the pre-growth period, t < T .
26
This is a non-stationary problem. However, given that the solution technique developed
in section 2 is applicable to non-stationary problems as well as stationary, this is straight
forward. We implement the dynamic program of Section 2, but working backwards in time
from the final condition Jq,GRW (ρ, C, V ), at time T .
4.1.3 The Optimal Policy with Growth Opportunities
To implement the model with growth opportunities we assume that prior to growth the firm
has exactly the same technology, capital structure, and cost parameters as in the benchmark
solution described in section 2.4. We have implemented growth for a variety of choices of
the pre-growth scale, V and the maximum post-growth scale, Vmax. The scale and equity
increments are set at dV = 0.001 and dQ = 0.001.
The solution to the growth problem is characterized by an investment threshold (ρg(C), C)
such that for a given level of cash holding C, there is a minimum rate of profitability given
by, ρg(C) below which the firm will not invest. The amount of investment chosen, given C,
is a weakly increasing function of ρ up to some point where Vmax is attained.
The investment threshold depends crucially upon the cost of investment as reflected in
the parameter V . For V = 0.10 the threshold is something less than ρ = 0.15 for most C.
To understand why note that in Table I, Panel A in the benchmark case of q = 0.004 if
ρ = 0.15 the value of the firm is about 0.11. Thus when V = 0.10 and ρ = 0.15, expanding
the scale of firm will be a positive NPV project if there are sufficient internal funds available
or if equity issuance costs are not excessive. Since ρ = 0.15 this is highly likely to occur.
In contrast, if V = 0.15, the cost of investment is relatively high and will be worthwhile
only at relatively high levels of profitability. Specifically the investment threshold for most
values of C is in the neighborhood of ρ = 0.20. Again this can be understood by reference to
Table I, Panel A, column q = 0.004 which results in a firm value somewhat in excess of 0.15 if
ρ = 0.20. Since ρ = 0.15 this occurs relatively infrequently. This case corresponds closely to
the typical pattern of marginal growth of capacity undertaken by relatively profitable firms
as documented in Table IV and discussed in section 4.1.1. Consequently, we use the value
V = 0.15 in the remainder of this section.
We turn now to the firm’s dividend and cash holding policy prior to the arrival of the
growth opportunity at t = T . In general the effect of the growth opportunity is to increase
the cash target for values of ρ where the investment will be a positive NPV project. For
lower values of ρ the growth option will have little effect on the cash target. The amount
by which the cash target is increased is an increasing function of the maximum scale of the
growth opportunity. That is, the greater the size of the likely investment, the greater the
intended accumulation of cash.
27
How does the cash holding policy evolve as the date of the investment opportunity ap-
proaches? We might expect that the the cash target is a decreasing function of the time
remaining until the arrival of the growth opportunity. For, if the cash reserve needed to
undertake the investment can be built up in a few months prior to the investment date then
there is no need to hold excessive amounts of cash very early on. However, this issue is
complicated by the fact that the optimal policy needs to take into account the likely evolu-
tion of profitability in the months prior to investment. When current profitability is high,
because of mean reversion it is likely that profitability will decline as the investment hori-
zon approaches. For this reason, at high levels of profitability it may be optimal to target
a higher level of cash accumulation when the investment horizon is relatively distant than
when it is close.
These qualitative features of the optimal cash holding policy are illustrated in Figure 7
for two sizes of potential growth and three investment horizons. We show the target 1, 3 and
6 months prior to the growth date T . The large opportunity is 30%, i.e., Vmax = 1.3V and is
depicted in the top figure. The small opportunity is an order of magnitude less,Vmax = 1.03V
and is given in the bottom figure.
For ρ less than about 0.19 all these targets are virtually identical and coincide with the
cash target for the no growth case depicted in Figure 1. The reason is that at these levels
of profitability it is unlikely that the firm will choose to undertake the opportunity when it
arrives at time T . Only for values of ρ in excess of 0.19 do the differences in the cash policies
emerge.
From the top figure (30% potential growth) we see that the impact of the growth op-
portunity is most dramatic in the case of a very large growth opportunity just prior to the
investment when the current profitability is high but not exceptionally high. In these cir-
cumstances the firm is attempting to accumulate as much as possible in order to undertake
what is likely to still be a positive NPV project once the investment opportunity arrives. At
very high levels of profitability the firm need not target such high levels of cash accumulation
since profitability is likely to remain high until the growth opportunity arrives giving the
firm ample time to build up the necessary internal funds. Note that for 30% growth when
ρ > 0.33 the cash target is higher at 3 months prior to investment than 1 month prior to
investment. This illustrates the property mentioned above that mean reversion implies that
at some time before the investment date it may be optimal to accumulate cash aggressively
when profitability is very high because profitability is likely to decline as the investment
horizon draws near.
Note also that these results on a large, lumpy investment opportunity imply that the
change in cash holding bears no simple relation to current cash flow, a point that has been
28
emphasized by Riddick and Whited (2009). In the run-up to the investment date high cash
flow will be associated with high cash accumulation. And the higher is profitability the
higher will be the proportion of cash cash flow retained because the higher profitability will
be associated with a larger planned investment. Later at the investment date high cash flow
is associated with sharp cash decumulation. Again the higher the profitability the higher
will be the rate of cash decumulation because the higher profitability is associated with a
larger investment.
The case of 3% growth illustrated in the bottom panel of Figure 7 is probably closer
to the typical firm’s investment plan as discussed in section 4.1.1. The presence of the
growth opportunity has no impact on cash targets for ρ < 0.19 and very little impact on
cash targets for higher levels of profitability. We saw that capacity expansion tends to be
undertaken by firms in the upper tail of profitability as is the case for our simulated firm
with V = 0.15. For this firm, at most times we will have ρ < 0.20, i.e., in a range where
the growth opportunity has little or no impact on cash holding. We have also calculated the
optimal policy for stochastic growth opportunity; in the case of a 30% opportunity which
occurs with 10% probability the results are very similar to the 3% opportunity just described.
In light of these results, we conclude that for a typical firm the anticipation of future growth
opportunities increases only slightly the observed levels of cash holding. This suggests that
the conclusions from the analysis of sections 2 and 3 above are robust to the inclusion of
growth opportunities.
4.2 Agency Conflicts and Debt Covenants
In our model, the liquidity reserve is chosen by the shareholders to maximize the share
value. This choice is made simultaneously with the default strategy, and the debt holders
are assumed to have no influence in these choices. It is thus possible that the liquidity reserve
strategy of the shareholders is detrimental to the debt holders and to the economic efficiency
of the firm.
In this section we will address a number of issues arising from this debt-equity conflict.
First we measure agency costs in our model. Specifically we ask what the consequences would
be for the debt, equity and firm values of maximizing firm value, rather than equity value.
The answer to this tells us the severity of the conflict in terms of how much the equity holders
are gaining from the bond holders via their liquidity strategy. We find that agency costs are
moderate in our model in good business conditions; that is, any loss to creditors of a share
value maximizing policy is compensated in large part by an increase in share value that is
of nearly equal magnitude. However, agency costs rise when business conditions deteriorate.
29
The reason is that costs of bankruptcy, either direct costs or indirect costs in the form of lost
tax shields, fall disproportionately on creditors. To gain more insight, we also investigate the
cases when the bankruptcy cost α is lower, at 5%, and τ = 0, so that there is no tax shield
to protect. When lower bankruptcy costs combine with zero tax shields, maximizing the
firm value rather than the equity value has a relatively small effect the firm value. Mostly
it is a pure transfer of value from shareholders to creditors. Generally, we find that the firm
value maximizing policy calls for the firm to hold more liquidity than under the share value
maximizing policy.
Then, in light of the sub-optimality of the share value maximizing liquidity policy, we
consider debt covenants which might be partial remedies. First we note that in view of the
result that firm value maximization calls for greater, not lesser, cash holding on average,
a covenant limiting the amount of cash held by the firm would be working in the wrong
direction. Instead, we consider a restriction on short-term borrowing and find that this leads
to very little increase in firm value. We then consider a second covenant prohibiting the
firm from paying a dividend when it is not profitable. This covenant is shown to increase
firm value. The merit of this covenant is that it curbs the firm’s ability to make the discrete
liquidating dividend. Eliminating this creates value for creditors because this helps to avoid
triggering bankruptcy prematurely.
4.2.1 Taking the Liquidity Reserve to Maximize the Firm Value
The levered firm value is the discounted expectation of dividends and payments to debt, up
to bankruptcy, at which time its value is 1 − α times its unlevered value. Following our
discussion in Subsection 2.3, if the firm value is being maximized, then its value F qt (ρ, C)
will satisfy
q +∂
∂tF q
t − rF qt + κ(ρ− ρ)
∂
∂ρF q
t +1
2ρη2 ∂2
∂ρ2F q
t
+ [(1− τ)(ρ− (f + q) + rCC)]∂
∂CF q
t +1
2σ2(1− τ)2 ∂2
∂C2F q
t = 0 (21)
in the ‘save’ region SF ;∂
∂CF q
t = 1, (22)
in the ‘dividend’ region DF ; and∂
∂CF q
t =1
θ. (23)
in the ‘issue’ region IF . Consistent with the boundary conditions in Subsection 2.3, if the
firm become bankrupt, then it has value (1−α)J00 (ρ, 0), and at the horizon time T , we take
30
the value to be C + min{q/r, (1− α)J00 (ρ, 0)}.
The numerical procedures described in the Appendix are easily adapted to this case,
and the result is summarized in Table V. In this table, the first two columns correspond
to our benchmark case; the first column repeats the relevant numbers from Table I. The
second column corresponds to maximizing the firm value. In this column, we calculate the
firm value as just described, and then the debt value as in Section 2.3, based on the regions
SF ,DF and IF , and the share value is then the firm value minus the debt value. We see, as
expected, that the net firm value (i.e. net of the value of the liquidity reserve) is higher, the
debt value his higher, and the net equity value is lower in the second column, than in the
first. Under normal or good business conditions agency costs are moderate. For example, at
ρ = 0.15 net firm value is 2.5% greater under the firm value maximizing policy. At ρ = 0.2
the agency cost is only 1%. However, at ρ = 0.1, which is close to the bankruptcy point
under the share value maximizing policy, the difference between firm value under the two
policies is substantial (9.6%).
It will be recalled that under the share value maximizing policy, at low values of ρ the
shareholders pay out available cash in the form of a discrete dividend and abandon the firm.
Under the firm value maximizing policy, the firm would retain its cash cushion as long as
possible, and it would keep the firm operating through equity issuance until ρ would fall to
a lower critical value. Overall, under the firm value maximizing policy, the firm holds larger,
not smaller, amounts of cash than under share value maximization. This casts doubt on the
wisdom of forcing shareholders to pay out cash as suggested in the quotation at the start of
this article.
Columns 3 and 4 repeat the case of columns 1 and 2, but with bankruptcy cost at 5%,
rather than 30%. We see that with low direct bankruptcy costs, the agency costs are less.
For example under normal business conditions (ρ = 0.15) firm value is greater by only
1.6% under the firm value maximizing policy (as compared to 2.5% with higher bankruptcy
costs). This is intuitive. Since the agency costs derive from the fact that shareholders impose
greater risks of bankruptcy on creditors than is optimal, when bankruptcy costs are less, the
associated agency costs are less.
The other cost of bankruptcy in our model is the indirect cost of tax shields lost in
liquidation. Thus in a similar vein, we repeat our experiment in columns 5 and 6, but with
τ = 0, so that there is no tax shield to be lost. As expected the agency costs of a share value
maximizing policy are less than under the benchmark parameters of α = 0.3 and τ = 0.3.
Finally, columns 7 and 8 present the case α = 0.05 and τ = 0. In this case, the agency costs
are quite small outside of financial distress. Furthermore, cash holding on average are quite
similar under the two policies.
31
4.2.2 Two Debt Covenants
In light of the agency costs incurred under the share-value maximizing liquidity policy, it is
interesting to consider whether simple, restrictive covenants might added to the firm’s debt
contracts as partial remedies.
Since the firm-value maximizing policy obtained in section 3A is a complex function
of the firm’s state variables, ρ and C, it may be difficult to come close to first-best using
covenants that are simple enough to be readily enforceable. Also, since we found that on
average firm-value maximization calls for holding greater, not lesser, amounts of cash than
shareholders would otherwise wish to hold, a covenant placing an upper limit on cash holding
is not likely to increase firm value.
This last observation suggests that a simple possible covenant might be to restrict the
amount cash position of the firm to be non-negative; that is, the firm would be prohibited
from borrowing short-term. The effect of this restriction can be seen in Table VI column
2 which can be compared to the benchmark solutions in column 1. Prohibiting short-term
borrowing has the effect of slightly raising the average cash holdings in the firm. However,
this covenant has negligible impact on firm the values of debt, equity, and the firm overall.
It is notable that it achieves virtually none of the 9.6% potential reduction of agency costs
in when the firm is near financial distress (ρ = 0.10). The ineffectiveness of this remedy is
attributable to the fact that a restriction on short-term debt impinges only for high values
of ρ, which a distressed firm is not likely to ever see. For the firm experiencing very good
business conditions (ρ = 0.2), the restriction on short-term borrowing slightly reduces net
firm value. That is, the gain to creditors does not compensate for the loss of share value.
It should be noted that this discussion assumes that the level of long-term debt is set at
our benchmark value, q = 0.004. However, in light of our analysis of the effect changing levels
of long-term debt as reflected in Figure 6, it is clear what the effect of a prohibition of short-
term borrowing would be for different levels of long-term debt. For higher q, unconstrained
cash holding becomes negative only for ρ in excess of ρ = 0.25 which is very unlikely to
occur. Thus the covenant would have virtually no effect on the firm. For the firm with less
long-term debt, (q < 0.004), a prohibition on short-term borrowing, could have an impact
on values. However, the main effect would be to prevent equity holders from benefitting
from tax shields under moderate and good business conditions. It would do little to protect
creditors when the firm is financially distressed. Thus on balance, the covenant is likely to
reduce firm value.
The problem with a simple prohibition of short-term borrowing is that since it is not
conditioned on the expected earnings of the firm, it is a very poor approximation to the state
contingent optimal policy. This same defect would carry over to other possible covenants
32
which would place a fixed lower bound on the firm’s cash position, e.g., Ct ≥ C where C
is a constant. The problem with this is that by setting the minimum cash holding at some
level that affords some protection to creditors it does so at the cost of lost earnings and tax
shields at high ρ.24
This discussion suggests that a more promising covenant design would condition upon
earnings and would seek to protect creditors when the firm is approaching financial distress.
In this regard we analyze a covenant which prohibits the firm from paying a dividend when
it is not currently sufficiently profitable. Leuz, Deller and Stubenrath (1998) note that debt
covenants restricting dividend payouts when the firm is in distress are common in the US,
but not in the UK. They argue that such covenants are considered unnecessary in the UK
because company law already restricts the conditions under which firms can pay dividends,
largely based on recent earnings.
In detail, we consider a covenant which prohibits dividend payments when ρt < f +q−L,
where L represents a measure of lee-way allowed under the covenant. This covenant will
automatically obviate the possibility of the discrete liquidating dividend, at least if L is not
chosen to be so big that the firm is allowed to pay dividends when it is not worthwhile
to continue operating. This covenant can be incorporated into our dynamic program, by
imposing Equation (6) above for all (ρ, C) such that ρ < f + q − L, irrespective of whether∂
∂CJq
t > 1.
The results are given in Table VI, Columns 3 and 4 corresponding respectively to L =
0.01, 0.02. Comparing these columns with Column 1, which corresponds to there being no
covenant, we see that the covenant does achieve some of the benefits sought. Under poor
business conditions (ρ = 0.1) firm value is about 2% greater than in the benchmark case
without the covenant. At higher values of ρ the net benefits are less, reflecting the fact that
the covenant would have effect only after a large fall of earnings which is relatively unlikely
to occur in the near future.
The enhancement of the debt at the expense of the equity might be attributed at least
partially to the direct value of the prospective liquidation dividend, which the covenant has
saved for the debt holders at the expense of the equity holders. However, the value of the
liquidating dividend is very small, when the firm is not currently in distress. This contingent
cash flow can be valued by simulating its discounted risk neutral expectation, and if currently
24It might be thought that this problem could be overcome by obliging the firm to hold the minimumcash in a trust account which would earn the money market rate r. This provision would mean that thefirm’s long-term debt would be equivalent to a portfolio consisting of a riskless secured bond equal to C plusa risky unsecured bond which is a claim on the flow (q − rC)dt. This is equivalent to simply reducing thelong-term debt of the firm, and the associated optimal policy would be the share-value maximizing policyfor that smaller debt. The net value of the firm would be unchanged.
33
ρ = 0.2, then its value is about 0.0000825. This is less than the wealth transfer between debt
and equity, associated with the covenant.
Preventing the transfer associated with the liquidating dividend cannot in any case ac-
count for the enhancement of the firm value, and this must be attributed to the fact that
the covenant makes bankruptcy less likely, since the firm will continue operating at any ρ
value, if there is some liquidity reserve.
It is interesting that the average amount of the liquidity reserve is actually slightly reduced
by this covenant and that the lee-way value L does not have much effect on the valuations.
Overall, our results are consistent with the view that covenants imposing a profitability
condition for dividends are value increasing.26
5 Conclusion
Using a structural dynamic model of the firm we have studied the optimal policy for liquid
asset holding, dividends and the issuance of both equity and short-term debt. We have
shown with this model that it is in shareholders’ interests to hold large amounts of cash
inside the firm even if they have access to relatively efficient capital markets and even if
some of the return to cash is dissipated by insiders. The reason that forcing the firm to pay
out “idle” cash in good times may be a bad idea is that it is myopic– it does not properly
weight the future dilution costs to shareholders of raising funds, if the firm later were to
approach financial distress.
We have shown that the firm’s policy toward liquid asset holding is closely connected to
the question of the firm’s capital structure. In particular we show that higher levels of long-
term debt will result in higher levels of liquid asset holding and a reduction in the optimal
use of short-term debt. The value of the firm is rather insensitive to the long-term debt level
outstanding. The reason is that by adapting its liquidity policy appropriately, the firm is
able to balance off its various contracting frictions in such a way as to achieve approximately
25Simulation also shows that the a liquidating dividend will be paid in about one quarter of bankruptcies,and if it is paid, then its average value will be about 0.01. The value at ρ = 0.2 is so low because at such ρ
bankruptcy is not expected for decades.26Something like the same effect might be achieved by placing a limit on the speed with which cash can
be paid out as dividends. In normal times the dividend as we have modelled it could be smoothed to havefinite speed with very little erosion of value. The value of such a speed limit when the firm is in distresswould be that the firm would likely have returned from distress by the time a substantial dividend were paid.There are a large number of ways such a restriction could be formulated; however, we believe that the orderof magnitude of the gain is likely to be close to that achieved through earnings restriction on dividends wehave studied here.
34
the same value of the firm for a wide range of long-term debt levels.
Our model exhibits realistic comparative statics behavior relative to a number of contract-
ing frictions including equity issuance costs and managerial rent extraction from financial
slack. We further explore the empirical implications of the model by simulating the model
in the baseline case calibrated to match simultaneously benchmarks for mean levels of cash
holdings, leverage, equity volatility, yield spreads, default probabilities and recovery rates.
Then using sample paths generated from this baseline model, we have replicated a number
of the cross-sectional and time series results reported in recent empirical studies of cash
holding. Our model also exhibits a state-dependent hierarchy among financial contracting
alternatives, and this is shown to be in line with results of recent tests of the pecking order
hypothesis.
When we extend the model to allow for increasing the scale of the fixed asset we find that
as expected the presence of growth opportunities tends to increase cash holding but that
this effect is significant only when the growth opportunity is large and imminent. For the
typical firm that grows marginally and only when current profitability is high, the presence
of growth options increases the level of cash holdings only slightly. The fact that cash may
be accumulated in anticipation of the arrival of a growth opportunity and then deccumulated
when the investment is realized implies that there is no stable relationship between cash flow
and rate of cash accumulation. That is, in agreement with recent work we find the rate of
corporate savings is highly variable and path dependent.
While the interests of shareholders and creditors are not perfectly aligned in our model,
they are not totally divergent either. In particular, we find that in most circumstance both
investor classes are harmed by increases in cash-flow volatility. Thus we cast doubt on the
generality of the intuition on the ‘asset substitution problem’ derived from static corporate
finance models. Furthermore, our comparison of a share-value-maximizing policy with a
firm-value-maximizing policy reveals that the targeted levels of cash are often quite similar
under both policies. The one area where their interests clearly diverge is when the firm’s
earnings are persistently low. In such circumstances, it is optimal for shareholders to pay out
available cash in the form of special dividend, thus exposing the firm to financial distress. In
this regard we have shown that it may enhance firm value to prevent the firm from paying
dividends when earnings are low, as is often done through bond covenants or as seen in some
systems of corporate law.
35
Appendix: Numerical Techniques and Boundary Condi-
tions for Solving the PDEs for Valuing Debt and Equity
Our strategy for valuing the equity is to solve Equation (6) numerically, by finite difference
procedures, evolving backwards from a distant horizon time t = T , at which we assume the
firm is liquidated. We have also changed the variable ρ to ξ := 12ρ
12 . This change helps the
numerical stability of the scheme because after we have expressed Equation (2) in terms of
η, the noise coefficient becomes constant. Also, under this transformation, the solution is
more detailed for low values of ρ, which are more important.
We use the explicit finite difference scheme (see Ames (1992)), representing the ρ space
by a grid with 201 points, ranging from 0 to ρmax = 3, and we represent C by a grid with
201 points ranging from Cmin = −0.2 to Cmax = 0.3. With these parameters, and for all the
parameters in the text, the finite difference scheme is numerically stable, if we take a time
step of length 0.01. Also, we take T = 50 years. By experimentation, we have determined
that the solutions are insensitive to first order variations of these parameters. To determine
the regions S, D and I, we test, at every time step and every grid point point representing
(ρ, C), whether it is optimal to pay dividends, issue shares, or abandon the firm.
Our numerical scheme for valuing the debt is the same as for the equity. The debt
boundary condition for bankruptcy is (1−α)J0(ρ, C), and this condition is calculated using
an implementation of the equity valuation for zero long term debt. Our calculations for the
credit spread, the probability of default at horizon 20 years, etc, are not based on evolution
to a steady state. These are obtained by taking T = 20 years, etc, in the above. Although
it might be more usual to use the technique of Successive Over-Relaxation (SOR) to obtain
our steady state solution (see Ames (1992)), our finite difference scheme is more useful for
studying how quickly the steady state is achieved, and for dealing with these non-steady
state calculations.
Our horizon boundary conditions for debt and equity at t = T are as follows: To value
unlevered equity for the valuations on bankruptcy, we take JunleveredT (ρ, C) = max{C, 0}, re-
flecting abandonment at time T . For the debt horizon value we take P qT (ρ, C) = min{q/r, (1−
α)Junlevered0 (ρ, 0) + C}, and for the equity horizon value we take Jq
T (ρ, C) = max{(1 −α)Junlevered
0 (ρ, 0) + C − q/r, 0}. These reflect the assumption that if bankruptcy has not
occurred by time T , the productive asset is sold, incurring the bankruptcy costs. The face
value of the debt, i.e.q/r, is paid out of the proceeds plus cash reserve to the extent possible,
and the rest goes to the equity holders. In section 3 A, when maximizing the firm value, we
take FT (ρ, C) = (1−α)Junlevered0 (ρ, 0)+C. Note that the choice of terminal time T condition
does not matter, if T is sufficiently far away, but a good choice allows T to be taken smaller,
36
which is more efficient, and this choice prevents arbitrage at time T .
As well as the boundary conditions associated with the regions S, D and I, we must
also choose boundary conditions at high and low values of ξ ≡ 12ρ
12 and C. The lowest
value of ρ is zero, and since dρ/dξ = 0, then any smooth boundary condition here for ∂∂ρ
J
will translate to ∂∂ξ
J = 0. We thus impose this condition. On the other hand, it is unclear
what the boundary behavior should be for high ξ. We therefore define a boundary region,
corresponding to ρ in the interval [ρmax, ρ+max) with ρ+
max = 3.5, and we extend the drift
coefficient of Equation (2) to be µ(ρ) = −2κρ+max−ρmax
πtan( π
2(ρ+max−ρmax)
(ρ − ρmax)) in this
region. This extends the drift coefficient smoothly past ρ = ρmax ,and has infinite slope at
ρ+max, and so the boundary condition ∂
∂ξJ = 0 holds. With the economic parameters given
in the text, the state variable ρ will very rarely rise as high as ρmax = 3, and so altering the
state equation above this value will not affect the solution in a measurable way. Our choice
of boundary region here has the value of preserving the numerical scheme, and ensuring that
no unwelcome singularities are introduced into the solution at the boundary.
37
TABLE I
Panel A - Values of Equity, Debt and LiquidityPayment to debt, q 0.000 0.001 0.002 0.003 0.004 0.005 0.006
Measures at ρ = 0.10Liquidity -0.0202 0.0144 -0.0041 0.0017 0.0026 0.0042 0.0058Equity val 0.0711 0.0555 0.0395 0.0232 0.0117 0.0049 0.0014Debt val 0.0000 0.0167 0.0333 0.0483 0.0562 0.0529 0.0468Net firm val 0.0711 0.0721 0.0728 0.0715 0.0679 0.0578 0.0481Leverage 0.0000 0.2888 0.4843 0.6597 0.7971 0.8537 0.8676Equity volatility 0.4659 0.5982 0.7062 0.9723 1.5381 1.8705 1.3699
Measures at ρ = 0.15Liquidity -0.0032 -0.0008 0.0040 0.0108 0.0136 0.0145 0.0138Equity val 0.1113 0.0968 0.0810 0.0652 0.0533 0.0441 0.0366Debt val 0.0000 0.0167 0.0333 0.0490 0.0609 0.0666 0.0708Net firm val 0.1113 0.1134 0.1144 0.1143 0.1142 0.1107 0.1074Leverage 0.0000 0.1480 0.2815 0.3920 0.4766 0.5318 0.5844Equity volatility 0.2396 0.2726 0.3104 0.34911 0.3962 0.4477 0.5138
Measures at ρ = 0.20Liquidity -0.0042 0.0008 0.0015 0.0025 0.0044 0.0064 0.0063Equity val 0.14827 0.1339 0.1184 0.1028 0.0908 0.0814 0.0735Debt val 0.0000 0.0167 0.0333 0.0491 0.0614 0.0679 0.0732Net firm val 0.1482 0.1505 0.1517 0.1519 0.1521 0.1493 0.1467Leverage 0.0000 0.1106 0.2174 0.3180 0.3920 0.4364 0.4781Equity volatility 0.2064 0.2214 0.2498 0.2861 0.3166 0.3425 0.3742
Measures at ρ = 0.25Liquidity -0.0211 -0.0131 -0.0099 -0.0034 -0.0020 0.0011 0.0031Equity val 0.1851 0.1709 0.1556 0.1401 0.1281 0.1186 0.1107Debt val 0.0000 0.0167 0.0333 0.0491 0.0616 0.0685 0.0742Net firm val 0.1851 0.1876 0.1889 0.1892 0.1897 0.1871 0.1848Leverage 0.0000 0.0955 0.1862 0.2644 0.3281 0.3639 0.3946Equity volatility 0.2024 0.2109 0.2291 0.2448 0.2655 0.2793 0.2931
38
TABLE I
Panel B - Credit Spreads and Bankruptcy ProbabilitiesPayment to debt, q 0.000 0.001 0.002 0.003 0.004 0.005 0.006
Measures at ρ = 0.10Credit spreads:
5 year PDB 0.0000 0.0000 0.0002 0.0072 0.0397 0.1355 0.295820 year PDB 0.0000 0.0000 0.0001 0.0034 0.0181 0.0577 0.1136
Prob. bankrupt after:1 year 0.0000 0.0000 0.0003 0.0144 0.0992 0.3516 0.67185 years 0.0000 0.0000 0.0005 0.0211 0.1276 0.4062 0.714820 years 0.0000 0.0000 0.0008 0.0301 0.1786 0.5105 0.7988
Measures at ρ = 0.15Credit spreads:
5 year PDB 0.0000 0.0000 0.0001 0.0016 0.0090 0.0274 0.048520 year PDB 0.0000 0.0000 0.0001 0.0020 0.0105 0.0307 0.05211
Prob. bankrupt after:1 year 0.0000 0.0000 0.0000 0.0001 0.0015 0.0060 0.01505 years 0.0000 0.0000 0.0001 0.0029 0.0184 0.0595 0.109820 years 0.0000 0.0000 0.0003 0.0129 0.0756 0.2244 0.3719
Measures at ρ = 0.20Credit spreads:
5 year PDB 0.0000 0.0000 0.0000 0.0011 0.0059 0.0185 0.033120 year PDB 0.0000 0.0000 0.0001 0.0019 0.0097 0.0284 0.0482
Prob. bankrupt after:1 year 0.0000 0.0000 0.0000 0.0000 0.0001 0.0003 0.00095 years 0.0000 0.0000 0.0000 0.0017 0.0109 0.0367 0.069120 years 0.0000 0.0000 0.0003 0.0116 0.0683 0.2051 0.3426
Measures at ρ = 0.25Credit spreads:
5 year PDB 0.0000 0.0000 0.0000 0.0008 0.0046 0.0147 0.026420 year PDB 0.0000 0.0000 0.0001 0.0018 0.0093 0.0274 0.0465
Prob. bankrupt after:1 year 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.00015 years 0.0000 0.0000 0.0000 0.0012 0.0081 0.0279 0.053320 years 0.0000 0.0000 0.0003 0.0111 0.0655 0.1974 0.3309
39
TABLE II: Comparison of Model Solution to Empirical Benchmarks
The empirical benchmarks are 1997-1999 median values for US industrial firms as reported by Standard andPoors for firms rated A, BBB and BB (Source, Standard and Poors, “Adjusted Key Ratios,” Credit WeekSeptember 20, 2000). The Model values are based on the simulation of the model under the benchmarkparameters, evaluated at the case ρ = 0.15. The model equivalent of EBIT interest coverage is computed as(ρ− f)/q. Return on capital is (ρ− f)/(P + J). The Long Term Debt to Capital ratio is P/(P + J). TheTotal Debt to Capital ratio is (P − C)/(P + J). Debt/Equity is (P − C)/J .
A BBB BB ModelEBIT interest coverage 6.8 3.9 2.3 2.5Return on capital 0.199 0.140 .117 0.15LT Debt/Capital 0.325 0.41 0.558 0.477Total Debt/Capital 0.401 0.474 0.613 0.370Debt/Equity 0.197 0.437 0.905 0.707
40
TA
BLE
III
-C
ompar
ativ
eSta
tics
Val
ues
ofE
quity,
Deb
tan
dLiq
uid
ity
Para
met
erbei
ng
vari
edN
one
ησ
κα
θr i
nr b
an
k
Para
met
ervalu
e(R
epea
t0.0
80.1
00.0
10.0
20.8
1.0
0.0
50.5
0.0
10.9
50.0
30.0
50.0
70.0
9
Ben
chm
ark
valu
eB
ench
mark
)0.0
90.0
90.0
00.0
00.9
0.9
0.3
0..3
0.8
00.8
00.0
40.0
40.0
80.0
8
TS
aver
age
liquid
ity
0.0
026
0.0
017
0.0
039
0.0
072
0.0
167
0.0
033
0.0
028
0.0
016
0.0
028
0.0
258
0.0
000
0.0
018
0.0
051
0.0
107
0.0
113
Mea
sure
sat
ρ=
0.1
0
Liq
uid
ity
0.0
026
0.0
017
0.0
039
0.0
072
0.0
167
0.0
033
0.0
028
0.0
016
0.0
028
0.0
258
0.0
000
0.0
018
0.0
051
0.0
026
0.0
028
Net
equity
val
0.0
117
0.0
123
0.0
113
0.0
106
0.0
084
0.0
089
0.0
150
0.0
133
0.0
117
0.0
042
0.0
258
0.0
109
0.0
130
0.0
117
0.0
117
Deb
tval
0.0
562
0.0
606
0.0
521
0.0
533
0.0
475
0.0
475
0.0
624
0.0
601
0.0
540
0.0
122
0.0
660
0.0
560
0.0
565
0.0
562
0.0
562
Net
firm
val
0.0
679
0.0
729
0.0
633
0.0
639
0.0
559
0.0
564
0.0
774
0.0
734
0.0
657
0.0
164
0.0
918
0.0
669
0.0
695
0.0
679
0.0
679
Lev
erage
0.7
972
0.8
121
0.7
739
0.7
494
0.6
546
0.7
964
0.7
775
0.8
009
0.7
881
0.2
897
0.7
186
0.8
152
0.7
565
0.7
972
0.7
946
Equity
vola
tility
1.5
382
1.4
514
1.5
490
1.3
667
1.1
971
1.7
950
1.1
951
1.5
220
1.5
120
0.5
312
0.8
217
1.7
123
1.2
263
1.5
384
1.5
137
Mea
sure
sat
ρ=
0.1
5
Liq
uid
ity
0.0
136
0.0
123
0.0
152
0.0
174
0.0
228
0.0
144
0.0
140
0.0
104
0.0
142
0.0
368
0.0
038
0.0
119
0.0
184
0.0
136
0.0
141
Net
equity
val
0.0
533
0.0
545
0.0
523
0.0
513
0.0
471
0.0
539
0.0
533
0.0
548
0.0
532
0.0
440
0.0
633
0.0
524
0.0
545
0.0
533
0.0
533
Deb
tval
0.0
609
0.0
639
0.0
575
0.0
594
0.0
560
0.0
559
0.0
644
0.0
631
0.0
596
0.0
339
0.0
663
0.0
608
0.0
612
0.0
609
0.0
609
Net
firm
val
0.1
142
0.1
184
0.1
098
0.1
107
0.1
032
0.1
098
0.1
178
0.1
179
0.1
128
0.0
778
0.1
296
0.1
132
0.1
157
0.1
142
0.1
142
Lev
erage
0.4
766
0.4
891
0.4
597
0.4
637
0.4
448
0.4
503
0.4
891
0.4
917
0.4
691
0.2
955
0.4
968
0.4
859
0.4
566
0.4
766
0.4
747
Equity
vola
tility
0.3
962
0.3
533
0.4
348
0.4
132
0.4
800
0.4
317
0.3
564
0.4
072
0.3
921
0.3
256
0.3
846
0.4
131
0.3
612
0.3
962
0.3
927
Mea
sure
sat
ρ=
0.2
0
Liq
uid
ity
0.0
044
0.0
029
0.0
072
0.0
092
0.0
160
0.0
046
0.0
054
0.0
025
0.0
064
0.0
263
-0.0
001
0.0
041
0.0
083
0.0
044
0.0
056
Net
equity
val
0.0
908
0.0
921
0.0
898
0.0
887
0.0
846
0.0
957
0.0
873
0.0
923
0.0
907
0.0
813
0.1
002
0.0
900
0.0
918
0.0
908
0.0
907
Deb
tval
0.0
614
0.0
642
0.0
582
0.0
600
0.0
569
0.0
569
0.0
646
0.0
634
0.0
602
0.0
360
0.0
663
0.0
613
0.0
616
0.0
614
0.0
614
Net
firm
val
0.1
521
0.1
563
0.1
480
0.1
487
0.1
415
0.1
525
0.1
519
0.1
557
0.1
509
0.1
173
0.1
665
0.1
512
0.1
534
0.1
521
0.1
521
Lev
erage
0.3
920
0.4
032
0.3
750
0.3
799
0.3
615
0.3
618
0.4
106
0.4
006
0.3
827
0.2
510
0.3
983
0.3
943
0.3
812
0.3
920
0.3
891
Equity
vola
tility
0.3
166
0.2
824
0.3
444
0.3
244
0.3
599
0.3
353
0.2
948
0.3
176
0.3
102
0.2
800
0.2
971
0.3
208
0.3
003
0.3
167
0.3
126
41
TABLE IV: Capital Expenditures, Share Issuance and LT Debt Issuance
This table reports annual average capital expenditures, share issuance and long-term debt issuance crosstablulated by earnings quintile and long-term debt level. The firms included are the Compustat universe ofnon-financial firms with total assets of at least 100 million 2000 dollars. All variables are normalized by totalassets at year end. The numbers reported are averages weighted by total assets (with total assets winsorizedat 1000 million 2000 dollars. The time period is 1978-2008.
Average Capital Expenditures
Profitability Quintile
Leverage category Lowest 2 3 4 Highest Total
Lowest .05453 .0416 .0485 .06216 .08177 .057712
2 .05425 .04905 .05889 .06769 .08699 .063374
3 .05137 .0513 .06185 .07468 .10139 .068118
4 .06219 .05261 .06522 .07958 .11698 .075316
5 .05999 .06084 .0716 .08975 .12627 .08169
Highest .10284 .07318 .07982 .09386 .12803 .095546
Total .064195 . 054763 .06431333 .077953 .1069 .073626
Average Share Issuance
Profitability Quintile
Leverage category Lowest 2 3 4 Highest Total
Lowest .11918 .03871 .01671 .00738 -.01204 .033988
2 .05108 .01271 .00762 -.00069 -.01491 .011162
3 .02019 .01025 .00776 .00215 -.00428 .007214
4 .03088 .01496 .01035 .00645 -.00387 .011754
5 .02466 .01443 .01247 .01326 .01204 .015372
Highest .05836 .01637 .01567 .01642 -.00023 .021318
Total .050725 .01790 .01176 .00750 -.00388 .016801
Average LT Debt Issuance
Profitability Quintile
Leverage category Lowest 2 3 4 Highest Total
Lowest .00507 .00612 .00832 .00555 .00282 .005576
2 .01585 .01783 .01717 .01168 .0043 .013366
3 .01238 .02061 .01708 .01125 .0063 .013524
4 .02691 .03149 .0258 .02097 .02007 .025048
5 .031 .0537 .04633 .02291 .00488 .031764
Highest .08061 .0739 .04154 .05206 .01872 .053366
Total .02864 .03394 .02604 .02073 .009515 .023774
42
TABLE V
Maximizing Equity versus Maximizing the Firm ValueTax rate τ 30% 30% 0% 0%Bankruptcy cost α 30% 5% 30% 5%Value being maximized Equity Firm Equity Firm Equity Firm Equity Firm
Measures at ρ = 0.10Liquidity 0.00261 0.00267 0.00161 0.00194 0.00723 0.00774 -0.00131 -0.00107Net equity val 0.01168 0.00793 0.01332 0.01093 0.02251 0.01985 0.02498 0.02240Debt val 0.05619 0.06652 0.06009 0.06662 0.06080 0.06662 0.06188 0.06663Net firm val 0.06787 0.07445 0.07340 0.07755 0.08331 0.08647 0.08686 0.08903Leverage 0.79724 0.86252 0.80106 0.83803 0.67154 0.70712 0.72334 0.75754Perp. cr. spread 0.01118 0.00013 0.00657 0.00004 0.00579 0.00004 0.00464 0.00003
Measures at ρ = 0.15Liquidity 0.01361 0.01402 0.01039 0.01162 0.02495 0.02641 0.01701 0.01757Net equity val 0.05328 0.05062 0.05479 0.05316 0.08130 0.07947 0.08436 0.08266Debt val 0.06090 0.06652 0.06307 0.06662 0.06353 0.06662 0.06388 0.06664Net firm val 0.11418 0.11713 0.11786 0.11978 0.14483 0.14609 0.14824 0.14930Leverage 0.47658 0.50719 0.49180 0.50700 0.37420 0.38618 0.38658 0.39937Perp. cr. spread 0.00568 0.00013 0.00342 0.00004 0.00296 0.00004 0.00261 0.00002
Measures at ρ = 0.20Liquidity 0.00444 0.00445 0.00254 0.00289 0.01194 0.01291 0.00678 0.00701Net equity val 0.09075 0.08818 0.09228 0.09070 0.13448 0.13271 0.13758 0.13596Debt val 0.06138 0.06652 0.06338 0.06662 0.06380 0.06662 0.06412 0.06665Net firm val 0.15213 0.15470 0.15566 0.15732 0.19827 0.19933 0.20169 0.20261Leverage 0.39203 0.41797 0.40060 0.41583 0.30349 0.31389 0.30755 0.31794Perp. cr. spread 0.00517 0.00013 0.00312 0.00004 0.00270 0.00004 0.00238 0.00002
43
TABLE VI
Analyzing Two Debt CovenantsBenchmark No S.T. borrowing No dividend if ρt < f + q − L
(No covenant) L = 0.01 L = 0.02
Measures at ρ = 0.10
Liquidity 0.002611 0.002849 0.000000 0.000000
Net equity val 0.011681 0.011722 0.010777 0.010779
Debt val 0.056193 0.056200 0.058564 0.058564
Net firm val 0.067874 0.067921 0.069341 0.069342
Leverage 0.797240 0.794120 0.844570 0.844560
Perp. cr. spread 0.011183 0.011175 0.008302 0.008302
Equit volatility 1.538630 1.508820 2.061170 2.060960
Measures at ρ = 0.15
Liquidity 0.013605 0.014243 0.008310 0.012239
Net equity val 0.053281 0.053277 0.052917 0.053083
Debt val 0.060900 0.060909 0.062254 0.062269
Net firm val 0.114180 0.114186 0.115170 0.115352
Leverage 0.476580 0.474260 0.504160 0.488040
Perp. cr. spread 0.005682 0.005672 0.004253 0.004237
Equity volatility 0.396230 0.391850 0.440210 0.407390
Measures at ρ = 0.20
Liquidity 0.004438 0.006416 0.005624 0.005635
Net equity val 0.090750 0.090720 0.090605 0.090606
Debt val 0.061379 0.061388 0.062636 0.062636
Net firm val 0.152129 0.152108 0.153241 0.153243
Leverage 0.392030 0.387250 0.394270 0.394240
Perp. cr. spread 0.005169 0.005160 0.003861 0.003861
Equity volatility 0.316640 0.309990 0.313230 0.313190
44
Figure 1:
This figure depicts the regions of ‘save’ (i.e. do retain earnings, and do not pay dividends),‘pay dividends’ and ’issue equity’, arising from the solution to the PDEs of Section 1.The ‘save’ region S is depicted by diamonds. The firm pays dividends for (ρ, C) not inthe ‘save’ region, so long as C is positive. The target liquid asset holding is the upperboundary of the S region and is depicted by downward triangles. The region in whichthe firm will issue new equity is indicated by upward triangles. These also indicate thelower limit of liquidity, or equivalently minus the limit of the short term borrowing fa-cility. The points where shareholders would pay a liquidating dividend and abandon thefirm to creditors are indicated with circles. The realized liquid asset holding, averagedover 300 simulations of the firm history, as a function of ρ, is depicted by the solid line.
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4−0.12
−0.1
−0.08
−0.06
−0.04
−0.02
0
0.02
0.04
0.06
0.08
Expected Revenue
Cas
h/S
.T.B
orro
win
g
save regioncash targetissue equityabandonaverage cash
45
Figure 2:
Simulated time series of the expected revenues ρt, (top graph), and the liquidity reserve Ct,equity value Jt, and total firm value Jt + Pt (all on the bottom graph, in increasing order).
117 118 119 120 121 122 123 124 125 126 1270.08
0.1
0.12
0.14
0.16
0.18
0.2
0.22
Year
Rev
enue
s
117 118 119 120 121 122 123 124 125 126 127−0.02
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
Year
Val
ues
cashequityfirm
46
Figure 3:
This is the distribution of coefficients from the regression of changes of the cash/asset ratioon its one-year lagged value as in Opler et.al. Figure 3. This is based on 121 years ofsimulated history for the benchmark model broken into sub samples of 5 years each. Asin the Opler et.al. study the coefficients are generally negative reflecting mean reversion.However, large positive coefficients are not uncommon, reflecting a reaction to a substantialchange of firm profitability which induces a persistent change in cash holdings. The mediancoefficient is -0.205 as compared to -0.242 found by Opler et.al..
−1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 10
1
2
3
4
5
6
7
Regression Model Coefficients
Num
ber
47
Figure 4:
Here the circles represent the value of the equity in the benchmark example, when theliquidity reserve is zero. The solid line is the value of collateral, i.e., 1 − α times unleveredfirm value. The diamonds represent the debt value in the benchmark case, for all levels ofliquidity.
0 0.05 0.1 0.15 0.2 0.25 0.3 0.350
0.05
0.1
0.15
0.2
0.25
expected revenues
valu
es
equityl.t.debtcollateral
48
Figure 5:
The top panel is the annual time series of the leverage defined as the book value of totaldebt to the value of the firm (book value of debt plus market value of equity) derived fromthe simulation of the benchmark model under the optimal policy. The lower panel is theplot of leverage against the previous year’s expected revenues. The estimated slope of theregression of leverage on lagged expected revenues is −2.6 and is significant at the 1% level.
0 20 40 60 80 100 120 1400.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Year
Leve
vera
ge
0.1 0.12 0.14 0.16 0.18 0.2 0.22 0.240.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Expected Revenues
Leve
rage
y = − 2.6*x + 0.85
49
Figure 6:
This figure depicts optimal policy for two different levels of long-term debt. The ‘x’ corre-spond to solution for the case q = 0.004. From top to bottom the lines give the cash target,the average cash and the debt-capacity/equity issue boundary and reproduce the results ofFigure 1. This is the “high debt” case. The lines in circles give the comparable results forthe “low debt” case, which is calculated for q = 0.0 i.e., no long-term debt.
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35−0.15
−0.1
−0.05
0
0.05
0.1
Expected Revenue
Cas
h/S
.T. B
orro
win
g
50
Figure 7:
These figures depicts optimal cash targets at 1, 3 and 6 months prior to growth for the firmwith either a 3% or a 30% growth opportunity.
0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4−0.04
−0.03
−0.02
−0.01
0
0.01
0.02
0.03
0.04
0.05
0.06
30% 1 mo30% 3 mo30% 6 mo
0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4−0.06
−0.04
−0.02
0
0.02
0.04
0.06
3% 1mo3% 3mo3% 6mo
51
References
Ames, W. F., 1992, Numerical Methods for Partial Differential Equations, Academic Press.
Anderson, R. W., 2003, Capital Structure, Firm Liquidity and Growth, in P.Butzen and C.Fuss (eds.),Firms’ Investment and Finance Decisions, Edward Elgar.
Anderson, R.W. and S. Sundaresan, 1996, Design and Valuation of Debt Contracts, Review ofFinancial Studies, Vol 9, pp 37-68.
Andrade, G. and S.N. Kaplan, 1998, How Costly Is Financial (Non-Economic) Distress? Evidencefrom Highly Leveraged Transactions that Became Distressed, Journal of Finance, Vol 53, pp 1443-1493.
Arrow, K., 1968 Optimal Capital Policy with Irreversible Investment, in J.N.Wolfe (ed.) Value, Capitaland Growth, Aldine.
Bolton, P. and X. Freixas 2000 Equity, Bonds and Bank Debt: Capital Structure and Financial MarketEquilibrium under Asymmetric Information, Journal of Political Economy, Vol 108, pp 324-351.
De Marzo, P. and Y.Sannikov, 2006, Optimal Security Design and Dynamic Capital Structure in aContinuous-Time Agency Model, Journal of Finance, Vol 61, pp 2681-2724.
de Servigny, A. and O. Renault, 2004, Measuring and Managing Credit Risk. McGraw-Hill.
Dittmar, A., J. Mahrt-Smith, and H. Servaes, 2003, International Corporate Governance andCorporate Cash Holdings, Journal of Financial and Quantitative Analysis, Vol. 38, 111-133.
Duffie, D., 2001, Dynamic Asset Pricing Theory, 3rd.ed. Princeton, Princeton University Press.
Duffie, D. and D. Lando, 2001, Term Structure of Credit Spreads with Incomplete Accounting Infor-mation, Econometrica, Vol 69, pp 633 - 664.
Duffie, D. and K.J. Singleton, 1999, Modeling Term Structures of Defaultable Bonds, Review ofFinancial Studies, Vol 12, pp 687 - 720.
Franks, J. and W. Torous, 1989, An Empirical Investigation of U.S. Firms in Reorganization, Journalof Finance, Vol 44, 747-769.
Ferreira, M. A., and A. Vilela, 2002, Why Do Firms Hold Cash? Evidence from EMU Countries,Working Paper ISCTE, presented at the European Finance Association Meetings, August 2003.
Fischer, E.O., R. Heinkel, and J. Zechner, 1989, Dynamic Capital Structure Choice: Theory andTests, Journal of Finance, Vol 44, pp 19-40.
Gomes, J., 2001 Financing Investment, American Economic Review, 91, 1263-1285.
Green, R, 1984, Investment Incentives, Debt and Warrants, Journal of Finance, Vol. 39, 115-136.
Hennessy, C. A. and T. M. Whited 2005 Debt Dynamics, Journal of Finance, 60, 1129-1164.
Kim C.-S., D. C. Mauer, and A. E. Sherman, 1998, The determinants of corporate liquidity: Theoryand Evidence, Journal of Financial and Quantitative Analysis , Vol 33, pp 305-334.
Leary, M.T. and M.R Roberts, 2006, The Pecking Order, Debt Capacity and Information Asymmetry.(February) Working Paper, Wharton School, forthcoming in Journal of Financial Economics.
Lee, I., S. Lockhead, J. Ritter, and Q. Zhao, 1996, The Costs of Raising Capital, Journal of FinancialResearch, Vol 1.
52
Leland, H., 1994, Risky Debt, Bond Covenants and Optimal Capital Structure, Journal of Finance, Vol49, pp 1213-1252.
Leland, H., 1998 Agency Costs, Risk Management and Capital Structure, Journal of Finance. 53,1213-1243.
Leuz, C., D. Deller, and M. Stubenrath, 1998, An International Comparison of Accounting-BasedPayout Restrictions in the United States, United Kingdom and Germany, Accounting and BusinessResearch, Vol 28, pp 111-129.
Mello, A. and J. Parsons, 2000, Hedging and Liquidity, Review of Financial Studies, Vol 13, pp 127-153.
Myers, S. and R.G. Rajan, 1998, The Paradox of Liquidity, Quarterly Journal of Economics, Vol 5,pp 733-771.
Nadiri, M.I. and I.R. Prucha, 1996 Estimation of the Depreciation Rate of Physical and RD Capitalin the U.S. Total Manufacturing Sector, Economic Inquiry, Vol. 34, pp 43-56.
Opler, T., L. Pinkowitz, R. Stulz, and R. Williamson, 1999, The determinants and implications ofcorporate cash holdings, Journal of Financial Economics, Vol 52, pp 3-46.
Petersen, M.A. and R. Rajan 1994 The Benefits of Lending Relationships: Evidence from SmallBusiness Data, Journal of Finance. Vol 49, 3-37.
Rauh, J. 2009 Risk Shifting versus Risk Management: Investment Policy in Corporate Pensions Plans,Review of Financial Studies Vol. 22, 2687-2734.
Riddick, L.A. and T.M Whited 2009 The Corporate Propensity to Save, Journal of Finance Vol. 64,1729-1766.
Rochet, J.-C. and S. Villeneuve, 2004, Liquidity Risk and Corporate Demand for Hedging and Insur-ance, working paper, GREMAQ, Toulouse University.
Shyam-Sunder, L and S. Myers, 1999, Testing Static Trade-off Against Pecking Order Models ofCapital Structure. Journal of Financial Economics. Vol 51, 219-244.
Smith, C. W., 1977, Alternative Ways of Raising Capital: Rights Versus Underwritten Offerings, Journalof Financial Economics. Vol 5, 273-307.
Strebulaev, I. 2007 Do Tests of Capital Structure Theory Mean What They Say? Journal of Finance,Vol 62, 1747-1787.
Warner, J., 1977, Bankruptcy Costs: Some Evidence, Journal of Financial Economics, Vol 5, pp 337-347.
Weiss, L., 1990, Bankruptcy Resolution: Direct Costs and the Violation of Priority of Claims, Journalof Financial Economics, Vol 27, pp 419-444.
Zhang,B.Y., H.Zhou, and H.Zhu, 2005, Explaining Credit Default Swaps Spreads with Equity Volatil-ity and Jump Risks of Individual Firms, BIS Working Paper.
53